Bio -based and A ppl ied Economics BAE Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6e172 | DOI: 10.36253/bae-12912 Copyright: © 2022 R. Esposti. Open access, article published by Firenze University Press under CC-BY-4.0 License. Firenze University Press | www.fupress.com/bae Citation: R. Esposti (2022). The co-evo- lution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period. Bio-based and Applied Economics 11(3): 231-264. doi: 10.36253/bae-12912 Received: March 17, 2022 Accepted: September 7, 2022 Published: November 4, 2022 Data Availability Statement: All rel- evant data are within the paper and its Supporting Information files. Competing Interests: The Author(s) declare(s) no conflict of interest. Editor: Davide Menozzi. ORCID RE: 0000-0002-1656-0331 The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Roberto Esposti Department of Economics and Social Sciences - Università Politecnica delle Marche, Ancona, Italy. E-mail: r.esposti@staff.univpm.it Abstract. This paper investigates the co-evolution of the CAP expenditure and of the farms’ performance and choices to assess whether and to what extent CAP assessment itself meets the requisites of Causal Inference. In order to identify some regularities in this co-evolution, the analysis is performed on a constant group of professional farms over a long enough time period. The Italian 2008-2019 FADN balanced sample is here considered. Results points to two major empirical implications. First of all, they ques- tion whether CAP expenditure is actually accompanied by any significant farmers’ response. An exception may actually concern the support specifically focused on envi- ronmental standards. Secondly, they raises some major methodological issues about the applicability of the Treatment Effect logic to CAP assessment. Keywords: Common Agricultural Policy, Farmers’ Behaviour, Program Evaluation, Panel Data, Co-evolution. JEL Codes: Q18, D04. “Verum scire est scire per causas” 1. INTRODUCTION: TWO TOPICS, ONE OBJECTIVE This paper deals with two distinct research topics and aims to join them into a unique research objective. The first topic consists in analysing the evolution of the Common Agricultural Policy (CAP) support, of the farm- ers’ production choices and of their possible interdependence (henceforth, the co-evolution). The second topic has to do with the growing use of the so- called Program Evaluation Methods (PEM) (Imbens and Wooldridge, 2009) in assessing the impact of the CAP, its measures and reforms, on the farming activity (Dumangane et al., 2021). The research objective that brings these two topics together is understanding whether and under which conditions investigating the farms’ response to CAP support can be performed with the cause-effect logic implied by these PEM. PEM have progressively emerged as the application of the general prin- ciples of Causal Inference (CI) to the assessment of public policies (Imbens http://creativecommons.org/licenses/by/4.0/legalcode 232 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 and Rubin, 2015; Perraillon et al., 2022). These methods are thus grounded on sound statistical concepts but, at the same time, they imply specific preconditions for an appropriate application to policy assessment (Khagram and Thomas, 2010). The bottom line is that an unam- biguous cause-effect direction must occur between a well-defined policy measure (the Treatment) and a well- defined response (the Treatment Effect, or TE). Such a direction (TE logic, henceforth) can be obvi- ously assumed but it is not necessarily a good represen- tation of the world especially in the case of many CAP measures. In particular, a correlation between some CAP measures (or reforms) and farmers’ behaviour does not automatically make the latter a response to the pol- icy. Not only because, as well known, correlation is not causation (Angrist and Pischke, 2009). More important- ly, as stressed by the literature on the political economy applied to the CAP decision making process (Swinnen, 2015; Collantes, 2020), a potential endogeneity may occur within this process. The main aftermath of such endogenous relationship is that CAP and farmer’s behav- iour rather co-evolve, so the observed correlation might express a cause-effect relationship whose direction, in fact, is not clearly identifiable. It follows that this paper is an empirical work but it is not an empirical application of some PEM to some CAP assessment. The empirical analysis rather aims to investigate the extent and nature of the abovemen- tioned co-evolution in order to assess whether and how it is compatible with the application of the TE logic. The main research question underlying this study is thus the following: which empirical support do we really have to interpret farmers’ behaviour as a response to CAP meas- ures and, thus, to consistently and properly apply the TE logic to CAP assessment? To answer these questions, the invariance of the field of investigation must be granted: a constant group (i.e., a balanced panel) of heterogeneous enough professional farms followed in its evolution over time together with the different CAP support they are recipients of. The Farm Accountancy Data Network (FADN) is helpful to perform this investigation, particularly in the Italian case where the FADN-RICA dataset contains most of the required information for the present analysis (Cagliero et al., 2010). Moreover, Italy presents a very diverse agriculture, and it is often considered the most heterogenous agriculture within the EU (Baldoni et al., 2021). Therefore, the 2008- 2019 Italian FADN balanced panel is here used. The abovementioned logic of the study also justi- fies its structure. Section 2 overviews the literature and the policy relevance underlying the present empirical investigation. Section 3 presents and discusses the bal- anced panel used for the analysis. Sections 4 examines the evolution of both CAP support and farms’ produc- tion choices and performance. Then, section 5 presents the co-evolution hypothesis by connecting these two dynamics and wondering to what extent one can be considered a response to the other. Section 6 derives the main consequences of this co-evolution in terms of the methodological challenges in adopting the TE logic in this field. Section 7 concludes drawing some methodo- logical implications. 2. THE POLICY ISSUE With the EU approaching the first year of applica- tion of its n-th CAP reform, expected to enter into force in 2023, the debate among agricultural economists, pol- icy experts and analysts remains essentially the same of the previous reforms. Positions range between two extremes. On the one hand, those (and the EU Commis- sion itself) who support the idea that this reform, as the previous ones, contain substantial novelties and some- how radical changes (European Commission, 2021; Pupo D’Andrea, 2021). On the other hand, others consider it, as the previous ones, essentially a conservation of the same fundamental schemes (same money, same ben- eficiaries, same modalities,) with only marginal or “cos- metic” changes (ARC2020, 2020; Sotte, 2021a). A sort of “conservative revolution”. What is common between these two opposite views is that both see the CAP as a policy expected to pro- duce an effect on (or a response by) the farming sector (OECD, 2011; Matthews, 2021).1 Maybe, however, this is not the proper perspective from which the CAP and its reforms have to be evaluated. The very fundamen- tal question is to what extent the CAP really condi- tions farmers’ choices and, therefore, whether it is really worth to adopt a TE logic (Coderoni, Esposti and Var- acca, 2021). In particular, the CAP presents three major problematic features in this respect. First, CAP is a policy and not a program, that is, is made of a set of interdependent measures (Lassance, 2020). These may be separately assessed (Castaño et al., 2019) but are not, usually, separately delivered to benefi- ciaries; and beneficiaries know this. In order words, the CAP is not a treatment, but it is a farm-specific (thus heterogeneous) combination of multiple treatments. Consequently, also the evaluation of individual measures 1 “Agricultural economists have been more concerned with the how and how well food and agricultural policies should be designed to achieve specific objectives and how policies have succeeded in their aims” (Mat- thews, 2021, p. 185-186). 233The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 should be performed only within a complex multiple- treatment environment. Secondly, the CAP is not just a set of measures, but it is a menu of measures since bene- ficiaries (farmers) are not assigned to some measures but voluntarily select among them (Esposti, 2022).2 Thirdly, this policy being a menu of measures, it turns out (in fact, it aims) to be a “passive” policy in the sense that is tailored on the existent rather than on inducing a change or a behavioural response. “Active” measures are present, but they may take the form of conditionalities, that is, requirements to be met in order to be eligible to a support. These conditionalities are usually quite weak, if not actually purely apparent, in the sense that most beneficiaries already satisfy them or need just minimum adjustments to satisfy them (Latacz- Lohmann et al., 2019).3 The key point here is that neither the CAP nor any CAP reform has a clear and univocal objective or target for which beneficiaries are expected to provide a specific response. CAP is a sort of “institutional environment” regularly accompanying, and not necessarily induc- ing, farms’ evolution. Eventually, the CAP behaves as a welfare system reserved to the EU farming sector. Its universalism (though limited to the farming activity) is expressed by the fact that its menu of measures covers nearly all farms, as well as all their different activities and instances.4 This does not exclude some more target- ed measures, but it remains true that multiple targeted measures ultimately aim to be universalistic. The main consequence of this universalism is that the CAP tends 2 The generalized voluntary nature of the CAP can be questioned. Here, voluntariness is intended in confront with the golden standard of ran- domized experiments where units assigned to the treatment do not choose whether or not to be treated. On the contrary, for all II Pillar measures the treatment is always the consequence of a voluntary choice. In the case of I Pillar direct payments, a difference has to be made between the period before and after 2015. After 2015, in practice all farms (but landless farms) have become entitled to apply for these pay- ments. Before 2015, those farms that did not receive coupled payments before 2005 were not entitled to apply and, therefore, could not vol- untary opt for the treatment. It remains true that, even when entitled, farms have to apply (so, to take a decision) and this also implies the respect of the cross-compliance conditions. Consequently, farmers that do not want to accept this conditionality may decide to do not apply even when entitled to do so. 3 There may be significant exceptions to this conclusion due to large het- erogeneity of agricultural systems across EU and Italy. For instance, in farming systems showing the prevalence of monoculture the introduc- tion of green payments, and the consequent compliance, had a relevant impact on farmers’ choices and behaviour (Bertoni et al., 2018; 2021). 4 This universalism does not conflict with the voluntary nature of most measures. It is rather the opposite: through a large set of voluntary measures, the CAP is able to provide assistance to all different kind of farmers according to their very different kinds of objectives. Voluntari- ness within universalism is, therefore, the obvious consequence of the large heterogeneity of beneficiaries. to be conservative and passive in the abovementioned sense. Rather than being one the effect of the other, the CAP and the farming sector seem to actually co-evolve.5 The nature of the CAP as an all-encompassing policy is not, per se, at odds with an evidence-based design and implementation (Esposti and Sotte, 2013; Erjavec  and  Erjavec, 2015; Erjavec, 2016; Ehlers et al., 2021). But this evidence concerns an expected effect (and, therefore, effectiveness and efficiency). Since this expected effect is unclear, the need of an evidence-based CAP inevitably raises the question: evidence about what? Waiting for the implementation of the new CAP reform (period 2023-2027), it seems useful to limit this ques- tion to the last 15 years. This is the period under inves- tigation here and it has been interested by two major reforms, implemented in 2005 and 2015, and by some major further adjustments meanwhile (particularly in 2007 and 2008). It can be argued that these reform steps share the same three fundamental objectives (Frascarelli, 2020, 2021; Coderoni et al., 2021): farm income support (or protection); farm competitiveness through (more) market orientation, i.e., (more) product diversification; larger and better public (mostly environmental) good provision by farms.6 In Italy, the decoupling of I Pillar support (the so- called Fischler Reform) was firstly introduced in 2005. It has been extended and reinforced in 2007 (with the introduction of the Single Common Market Organiza- tion, CMO) and in 2008 (the Health-Check Reform), and then progressively dissociated from historical direct pay- ments in 2015 (the Ciolos Reform) (Sotte, 2021b). Conse- quently, the period under consideration here (2008-2019) starts from a year in which the full decoupling of direct payments was already under way. Meanwhile, II Pillar support has been strengthened in terms of overall sup- port and of its share on the total CAP budget, but also 5 This is the empirical counterpart of the political economy argument on the endogeneity of the CAP (Swinnen et al., 2015) which suggests that its design may depend on farmers’ choices and behaviour more than the other way round. 6 Matthews (2021, pp. 185-191) overviews the evolution of the funda- mental objectives of the CAP over time. “Farm income”, “Environment” and “Competitiveness” are among the most persistent. The objective of production diversification and market reorientation can be considered an explicitation of the competitiveness objective. In fact, these are not the only objectives of the CAP but are those that directly and exclu- sively refer to farmers’ behaviour under scrutiny here. Other objec- tives could actually be added to this short list (European Commission, 2019; Coderoni et al., 2021). In particular, two are worth noticing. One is favouring structural change or adjustment within agriculture. The other is supporting the rural economy. But these objectives are beyond the horizon and, above all, the field of investigation of the present study both for the limited time under consideration and for the use of bal- anced panel of farms (see below) that, evidently, do not cover all socio- economic aspects of the rural economy. 234 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 in terms of a progressively stronger orientation towards environmental goods provision. With respect to the three abovementioned fundamen- tal objectives, the decoupling of support (with the main- tenance of the support level) was expected to induce mar- ket re-orientation while granting farmers’ income (Anton, 2006; Esposti, 2017a,b; Ciliberti et al., 2022). Also II Pil- lar had to facilitate market re-orientation (and structural change) and, at the same time, the environmental goods provision especially due to the strengthening of Agro- Environmental Measures (AEM) already introduced in the 1992 reform (MacSharry Reform). Pillar I itself has been designed to contribute to the environmental objec- tives with the introduction of the environmental condi- tionality already in 2005, then further enhanced with the novel Greening payments in 2015. Therefore, in principle, this sequence of reforms has been designed to get progres- sively closer to the abovementioned objectives. In practice, however, their actual implementation might not have gen- erated a major impact.7 A lot of research work has been done in order to directly investigate, simulate, estimate the impact of these CAP reform steps on beneficiaries. This large body of literature is definitely helpful in better understand- ing the mechanisms through which the CAP operates and, therefore, in better designating and implementing it (Matthews, 2021). But analysing the possible impact of the CAP and its reforms with these approaches does not nec- essarily correspond to a program evaluation. Most studies are grounded on farm-level structural models used either for ex-ante (simulations) or ex-post (simulations or esti- mations) assessment (see, for instance, Mack et al., 2019). Within their theoretical structure, these models somehow impose the existence, the form and sometime the direc- tion of the response to policy measures. Eventually, the problem is the lack of a counterfac- tual evidence. In most of these studies the counterfactu- als are never observed, and they might not even exist, but the counterfactual case is just extrapolated from the estimated models parameters. The search of such coun- terfactual evidence may explain the emergence, in the last fifteen years, of a consistent body of empirical stud- ies whose aim is to explicitly assess the CAP impact within a TE logic (just to mention a few: Chabé-Ferret and Subervie, 2013; Castaño et al., 2019; Coderoni, 7 Studies on the distribution of the CAP support across regions and farms (see Sotte, 2014, and Terluin and Verhoog, 2018, to mention a few) have mostly concluded that the beneficiaries and the allocation among them did not change significantly over time. This can be con- sidered an implicit demonstration that the (reform of the) CAP might not have had an effect. But this is not obvious. Maintaining the distribu- tion of support but changing the forms and modalities may still induce a response. Esposti and Varacca, 2021; Ciliberti et al., 2022; Esposti 2017a,b, 2022). This research effort is commendable and promising. As mentioned, however, the actual charac- teristics of the CAP and of its reforms do not necessar- ily fit the strict requirements of this TE logic. In most of these recent studies its suitability for CAP assessment is given for granted and never really questioned. In prin- ciple, preliminary to any TE investigation, it would be desirable to scrutinize the empirical support about the applicability of this logic to the three abovementioned key objectives. Looking for this empirical support is the main purpose of the present study. 3. THE DATA: 2008-2019 FADN ITALIAN BALANCED SAMPLE Another major issue in the investigation of farms’ responsiveness and co-evolution with respect to CAP measures concerns the field of investigation. Several pre- vious studies work on all farms, but this can introduce a bias as their response may be not fully observable for the presence of many very small farms (even “non-farms”) (Sotte, 2006; Sotte and Arzeni, 2013) and may be also driven by long-term structural processes that are largely independent on the CAP support. A further limitation of the field of investigation of many previous studies is the lack of a long-enough time dimension. Most of them are, in fact, ex ante assessments thus they are a-temporal in the sense that are based on current farm-level data possibly on the basis of future scenarios. They seldom take the needed time until the farms’ co-evolution or response is significantly revealed by data. Here, we focus on a sample that take these issues into account: the Italian 2008-2019 FADN balanced pan- el.8 A constant field of observation is clearly needed to investigate the co-evolution of the CAP support a farm- ers choice in order to get rid of the spurious effects sim- ply generated by the change in the sample composition. This choice, however, may also have limitations and two of them are worth noticing here. The first limitation is that working on the FADN sample may miss some of the 8 This balanced panel consists of 1585 farms observed over 12 years, thus 19020 total observations. Even if 2020 data were available, they are going to be problematic in terms of comparability due the effects of the COVID-19 pandemic also on the farming sector. The EU-wide FADN sample could be used instead but the information available over all countries are less comparable and, above all, less detailed than those reported in the Italian RICA-FADN dataset. The choice of working with a balanced panel also explains why some of the results here presented may also substantially diverge from what obtained in studies working on the same period but on a different fields of investigation (European Commission, 2019). 235The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 implications of CAP and its reforms as changes occur- ring in non-professional farms, numerically prevalent in the Italian context (Sotte, 2006; Sotte and Arzeni, 2013), remain unobserved as these units are excluded from the FADN field of survey. Structural changes may be also missing in the balanced panel. As the non-constant part of the FADN sample is excluded, the dynamics of entry/ exit (i.e. deactivation) from the sector, as well as other changes somehow related to the entry/exit from the sam- ple (for instance, change in size due to land acquisition or loss), are at least partially missed. However, none of this possibly missing information is at the core of the three CAP objectives here considered. The second limitation concerns the possible lack of representativeness of the adopted balanced panel with respect to whole Italian agriculture even when only professional farms are considered (Mari, 2020; Vrolijk and Poppe, 2021).9 Representativeness is evidently sac- rificed when a balanced panel is extracted over a long- enough period since the FADN sample is rotating just in order to maintain representativeness over time. It is thus informative to make explicit how much the adopt- ed dataset may over or under-represent some farms cat- egory compared to the whole Italian agriculture. Table A1 in the Annex compares the distribution of farms by Type of Farming (TF) and Economic Size class (ES) in the adopted sample (in year 2010) with the Italian 2010 agricultural Census.10 For the sake of comparison, Cen- sus data are reported in two forms: the whole farm pop- ulation and the population corresponding to the FADN field of survey, that is farms with a Standard Output (SO) higher than 8 thousand € (also called professional or market-oriented farms).11 It firstly emerges that the FADN sample always somehow misrepresents the whole Italian agriculture as about 63% of the farm population is excluded from the FADN field of survey. But limiting the attention to pro- fessional farms, the distribution of farms within the bal- anced FADN panel in terms of TF does not differ much from what observed in the Census data, even though a slight over-representation of grazing livestock activi- ties (TF4) and under-representation of permanent crop farms (TF4) is observed. A more important bias con- cerns the ES as the balanced panel evidently self-select larger farms, in economic terms. This bias has to be tak- en in mind in commenting the following results and any 9 We wish to thank two anonymous referees for their helpful suggestions and remarks on this aspect. 10 Together with the geographical district (regions in Italy), these are the two levels for which the representativeness of the FADN sample is granted (Mari, 2020). 11 In Italy, this threshold was 4 thousand € up to 2014. generalization to the whole Italian agriculture requires caution. However, it is worth stressing here that there is no feasible solution to this representativeness issue whenev- er a balanced FADN panel is adopted.12 Even the vector of individual weights that accompany the FADN sample cannot be helpful in this respect. These weights allow to carry over the sample-level evidence to the popula- tion, at least for those dimensions for which the FADN sample is representative (Mari, 2020). Therefore, weights are useful to compute population-level aggregates, given representativeness, but it is not suitable to recreate this representativeness. Moreover, these weights refer to the whole FADN sample and not to the balanced FADN sample. They also vary any year and have to be redefined any time the underlying sampling scheme is changed as occurred, in particular, with the change of the pro- fessional farm threshold in Italy in 2014 and with the change of the TF classification in 2010. Applying these weights to the balanced panel over 12 years would incur the risk of generating an uncontrollable distortion rather than correcting for an observed misrepresentation. Considering that working on a constant sample is a strict condition to properly investigate the co-evolution of CAP support and farms’ behaviour, we prefer here to sacrifice representativeness rather than to generate arte- facts in the attempt to correct for it. Also because repre- sentativeness is not a major concern with the respect of the major objective of the present paper. Evidently, any policy conclusion based on these data should be taken with major caution (Vrolijk and Poppe, 2021, p.10). But the main interest, here, is rather on the methodologi- cal implications of the co-evolution of CAP support and farms’ behaviour. It may be the case that such co-evolu- tion does not perfectly correspond to what observed in the whole Italian agriculture and may slightly overvalue the incidence of the outliers (in particular, farms with very high payments).13 However, evidence here reported remains valid within the adopted field of investigation and, more importantly, with respect to its main meth- odological implications. Within this sample, the empirical analysis is devel- oped in a sequence of three steps. First, the evolution of the CAP support and of its distribution is investigated, considering both its total amount and its components (section 4.1). Then, the evolution of the farmers’ choic- 12 In any case, it has been already noticed that also within the Italian FADN sample the full representativeness on the three abovementioned dimensions is more theoretical than actual (Mari, 2020, Tables 2 and 3). 13 In the present case, however, what could be considered outliers are actually real farms. They might be peculiar and, for this reason, they are recipients of a very high CAP support. But this does not mean that they represent anomalous or aberrant cases. 236 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 es and performance is analysed (section 4.2). Finally, some stylised facts about the co-evolution of these two dynamics are derived (section 5). 4. THE EVOLUTION OF THE CAP SUPPORT AND FARMS’ BEHAVIOUR 4.1. CAP support The first question to be answered is whether the CAP support actually changed within the adopted field of investigation and how. Figure 1 displays the total and per farm public support considering all the pos- sible sources.14 The total support remains quite regular over the period (always ranging between 23 and 27 mil- lion €) with only limited oscillations due to the transi- tion to one CAP regime to another. Overall, we observe an increase in total support (+17% from 2008 to 2019) in nominal terms, but this growth almost entirely vanish- es (+4%) in real terms (2010 prices).15 Consequently, the per farm average support passes from 14.4 thousand € to 16.9 thousand € per farm, in nominal terms. But in real terms this variation drops from 14.4 thousand € to 15.3 thousand € per farm. Figure 2 reports the evolution of the composition of the total CAP support. It evolved as a combination of three dynamics: 1. I Pillar declined by 4% and II Pillar grew by 156% and this has made the share of I Pillar and II Pillar be gradually re-equilibrated with the latter moving from a 13% to 29% of total CAP support. 2. Within I Pillar, decoupled support remained sta- ble (-0.4%) while coupled payments declined by -20% up to a final 15% in 2019 on total Pillar I payments (corre- sponding to 11% on total CAP support). The process of progressive decoupling of support actually stopped in 2012 since for the rest of period the shares of coupled and decoupled support remained quite stable. 3. Within II Pillar, the largest growth concerns AEM payments (+196%) while the other measures increased by 124% with AEM support passing from a share of 44% on the total II Pillar support in 2008 to 51% in 2019. The huge growth and the increasing rele- vance of the AEM support is investigated further in the Annex (Figure A1). 14 Regional co-financing of II Pillar is included in CAP support. The remaining national support represents a very marginal part, always low- er than 5%. For this reason, the national support will be neglected in the rest of the analysis. 15 Following Matthews (2000), real values are computed using the offi- cial Italian GDP deflator released by the National Institute of Statistics (ISTAT). The synthesis that can be drawn from this general picture is that, at least from the farms’ perspective, the evolution of CAP support in the 12 years under inves- tigation really represents a sort of “conservative revolu- tion”: the different components of the whole expenditure changed significantly, but the support eventually deliv- ered to farmers is more or less the same. Nonetheless, the key argument of the critics of this alleged conserva- tism of the CAP consists not so much in the amount of support but in its strongly uneven distribution across farmers. Table 1 reports some year-by-year distributional statistics of the total and CAP support, and of its differ- ent components, within the present sample. Overall, it is confirmed that values (but the maximum) are quite sta- ble over time. At the same time, the distribution is very disperse with a standard deviation always much higher than the mean value as indicated by a greater than two Coefficient of Variation (CV). Moreover, the left tail of the distribution being truncated at 0, the presence of several extreme values generates a remarkable asymme- try with a very long right tail. This is clearly revealed by the difference between the mean and the median (2nd quartile) values, with the former being in all cases more than double than the latter. High variability and asymmetry is observed in all the different policies but some specificities are worth noticing. In particular, both coupled I Pillar payments and non-AEM II Pillar payments show very high CV values. For both II Pillar subgroups the observed sup- port is zero until the third quartile indicating that pay- ments concentrate on a very limited number of farms.16 It can be also concluded, however, that these specific asymmetries tend to compensate, at least partially, as dispersion and asymmetry observed in the total support are significantly lower than in the single components. This apparent stability of the CAP support distribu- tion over time does not mean that from any individual farm perspective nothing changed. By looking at the single farm percentage variation of the received support from 2008 to 2019 (bottom of Table 1), it emerges that several farms lost all the support (-100%) while for oth- ers the growth is maximum (in fact, it can not be com- puted simply because the initial value is zero). Between these extreme cases, we find most farms with a change in the support that ranges from a decline (the first quar- tile is -19%) to a huge increase (the third quartile is +370%). The mean value (the second quartile) indicates a 16 A similar, in fact more extreme, case can be found in national pay- ments where also time variation is large. These distribution characteris- tics can be explained by the fact that national payments tend to have an emergency or exceptional nature: they are activated under very special conditions, for very specific farms and for a limited period of time. 237The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 30% growth which is consistent with the growth of aver- age support commented above. We should thus conclude that the evolution of the CAP over this period signifi- cantly redistributed the support across farms but did not make it more homogeneously distributed. 4.2. Farms’ behaviour 4.2.1. Profitability In order to assess whether or not this CAP evolu- tion had any relevant impact on farms’ performance and choices, the first question to be answered concerns farms’ profitability. Here we proxy the farm’s profit with the farm’s net income simply computed as revenue plus policy support less all costs.17 Therefore, in order to investigate the evolution of farms’ profitability it is 17 In the FADN terminology what is here referred to as Net Income cor- responds to the Entrepreneurial Income. As most agricultural produc- tion units are family farms, this also corresponds, for many units, to the Family Farm Income (European Commission, 2018a). The difference between net farm income and farm profit is that the former is defined as farm revenue, plus policy support, less all external costs; the latter as the difference between net farm income and the opportunity cost of factors of production (labour, land and capital) provided by the family farm. We wish to thank an anonymous referee for an helpful clarifica- tion on this point. worth to analyse the evolution of its components. Figure 3 displays the dynamics of the average revenue and vari- able costs within the field of investigation. A selection of these costs is also shown. They concern what we design here as environment-using costs: fertilizers, pesticides (herbicides included), energy and water. It firstly emerges a regular increase of both rev- enue and costs, but with the latter showing a larger growth than the former (+38% and +12%, respectively). It follows that the incidence of variable costs on revenue passes from 38% in 2008 to 47% in 2019. Among costs, environment-using ones maintain a quite constant share, always higher than 20% and lower than 25%. From these figures a quite regular profitability over the period can be deduced. Figure 4 shows that the average farm net income did not significantly change as it remains between 50 and 60 thousand €. A -10% variation is actually observed comparing 2019 with 2008, but this decline can be entirely attributed to the very last year. If we express net income in real terms, however, a different conclusion can be drawn. Although inflation has been constantly low during this period, in real terms the average farm net income suffered a -20% decline from 2008 to 2019 that becomes a -9% if we stop the comparison at 2018. We should thus rather conclude that, on average, farms actually struggled to defend their profitability over this period. At the same time, however, 0 2 4 6 8 10 12 14 16 18 20 0 5,000 10,000 15,000 20,000 25,000 30,000 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Total CAP support (000 €) - left scale Avg. per farm CAP support in nominal terms (000 €) - right scale Avg. per farm CAP support in real (2010 prices) terms (000 €) - right scale Figure 1. Total and per farm public support within the Italian 2008-2019 FADN balanced sample. 238 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 Figure 2. Composition of the total public (a) and CAP (b) support within the Italian 2008-2019 FADN balanced sample. 239The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 Table 1. Distribution of the public support (CAP included) within the Italian 2008-2019 FADN balanced sample (€). 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 TOTAL SUPPORT Mean 14,449 14,848 16,642 15,802 16,700 16,846 16,870 17,158 16,614 16,381 16,510 16,856 Standard deviation 31,844 31,258 41,230 38,001 39,722 38,967 42,899 42,205 41,005 40,174 36,552 34,645 Coefficient of Variation 2.2 2.1 2.5 2.4 2.4 2.3 2.5 2.5 2.5 2.5 2.2 2.1 Min 0 0 0 0 0 0 0 0 0 0 0 0 1st Quartile 755 1,014 1,666 1,860 1,924 1,844 1,904 1,508 1,501 1,532 1,804 2,325 2nd Quartile (Median) 5,065 5,498 5,920 6,341 6,545 6,536 6,329 6,470 5,941 6,185 6,449 6,826 3rd Quartile 14,824 15,904 17,100 16,878 17,634 17,607 17,431 18,625 17,807 16,971 17,813 18,489 Max 420,574 505,280 859,158 834,940 737,493 720,471 894,886 1,158,547 972,158 911,073 834,179 756,761 NATIONAL SUPPORT Mean 245 573 708 473 373 437 469 255 276 294 296 185 Standard deviation 1,883 4,874 8,621 4,410 3,126 3,396 4,983 2,219 2,408 2,233 1,974 2,379 Coefficient of Variation 7.7 8.5 12.2 9.3 8.4 7.8 10.6 8.7 8.7 7.6 6.7 12.9 Min 0 0 0 0 0 0 0 0 0 0 0 0 1st Quartile 0 0 0 0 0 0 0 0 0 0 0 0 2nd Quartile (Median) 0 0 0 0 0 0 0 0 0 0 0 0 3rd Quartile 0 0 0 0 0 0 0 0 0 0 0 0 Max 32,072 102,854 213,984 119,430 66,942 69,493 145,670 60,900 56,491 59,487 36,800 74,000 TOTAL CAP Mean 14,204 14,275 16,106 15,329 16,327 16,410 16,401 16,903 16,338 15,753 16,286 16,599 Standard deviation 31,763 30,850 38,511 37,724 39,512 38,714 42,579 42,135 40,897 37,516 36,501 34,463 Coefficient of Variation 2.2 2.2 2.4 2.5 2.4 2.4 2.6 2.5 2.5 2.4 2.2 2.1 Min 0 0 0 0 0 0 0 0 0 0 0 0 1st Quartile 724 947 1,622 1,840 1,892 1,750 1,814 1,491 1,461 1,497 1,770 2,286 2nd Quartile (Median) 4,957 5,144 6,395 6,125 6,478 6,325 6,130 6,340 5,727 5,955 6,257 6,701 3rd Quartile 14,611 15,320 17,745 15,969 17,132 17,157 16,961 18,337 17,253 16,514 17,417 18,245 Max 420,574 505,280 805,154 834,940 737,493 717,971 894,886 1,158,547 972,158 911,073 834,179 752,234 PILLAR I - DECOUPLED Mean 9,961 9,954 11,211 11,178 12,750 12,536 12,886 11,541 11,340 10,883 10,291 9,922 Standard deviation 21,774 21,001 33,082 31,119 35,614 34,153 36,599 32,735 30,054 26,924 24,078 21,307 Coefficient of Variation 2.2 2.1 3.0 2.8 2.8 2.7 2.8 2.8 2.7 2.5 2.3 2.1 Min 0 0 0 0 0 0 0 0 0 0 0 0 1st Quartile 98 215 507 768 886 830 878 739 925 1,028 1,204 1,242 2nd Quartile (Median) 3,477 3,483 4,015 4,199 4,231 4,171 4,068 3,568 3,614 3,644 3,682 3,830 3rd Quartile 10,715 10,992 11,718 11,772 12,412 12,095 12,179 11,040 11,008 10,519 10,445 10,378 Max 317,849 319,288 801,933 724,970 720,596 680,898 759,890 862,371 631,221 558,244 528,809 417,296 PILLAR I – COUPLED Mean 2,374 2,380 1,923 1,577 567 631 776 1,726 1,883 1,744 1,736 1,889 Standard deviation 11,566 12,675 8,401 8,003 3,357 3,786 5,307 8,515 9,999 10,320 8,860 9,275 Coefficient of Variation 4.9 5.3 4.4 5.1 5.9 6.0 6.8 4.9 5.3 4.9 5.1 4.9 Min 0 0 0 0 0 0 0 0 0 0 0 0 1st Quartile 0 0 0 0 0 0 0 0 0 0 0 0 2nd Quartile (Median) 0 0 0 0 0 0 0 0 0 0 0 0 3rd Quartile 690 794 0 0 0 0 0 975 1,087 1,034 1,221 1,120 Max 237,355 340,652 122,828 124,584 90,000 108,794 134,996 293,872 338,633 352,829 305,370 312,498 PILLAR II – AEM Mean 826 931 1,100 1,127 1,602 1,620 1,366 1,946 1,940 1,917 2,279 2,450 Standard deviation 2,991 3,300 4,132 4,430 5,390 5,585 5,043 6,552 6,674 6,555 7,528 7,862 240 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 the number of farms with negative net income did not increase. It amounted to 9% of the whole sample in 2008 and to 7% in 2019, and has remained always between 10% and 5% though with a clear drop after 2009.18 More generally, average values may be uninforma- tive, and even misleading, due to the large heterogene- ity occurring within the panel as also detailed in the Annex (Figure A3). Table 2 illustrates how during these twelve years the farm net income dispersion and asym- metry maintained the same basic features with no major evidence of a more uniform distribution. Such large dispersion is confirmed by a CV always around two or 18 A classical issue in the analysis of farm profitability concerns whether the farming activity can grant agricultural workers and families a com- parable income with respect to the rest of the economy. This issue has generated a long debate among agricultural economists, particularly in Italy (Rocchi et al., 2012). Present results may provide some indica- tion in this respect even though, as discussed, the net farm income here considered does not correspond, stricu sensu, to the family farm income for all units. In addition, as discussed, the adopted sample only consider commercial farms and tends to be biased upward, i.e., to have a little overrepresentation of larger farms in economic terms. Nonetheless, for the sake of comparison, it can be noticed that the mean net farm income in the last year of observation (51,440 €) is significantly higher than the average family income resulting, for the same year, from the Italian Sta- tistics on Income and Living Conditions (SILC) (33,653 €). This remains true even when only families with prevalent autonomous work are con- sidered (42,340 €). However, it should be also noticed that this positive gap can be a further consequence of the asymmetry within the sample. If the median net farm income is considered (23,154 €) the gap seems to be actually reversed. Moreover, while the average (or median) net farm income observed within the sample shows a decline in real terms over the period under analysis, the average real-term family income resulting from the SILC data show a very slight increase (+0.4%). more, though it also shows a decline in the last three years under observation. The same does not occur for the asymmetry that remains large and constant over the whole period, with a very long right tail that motivates why the mean value is always more than double than the median value (2nd quartile). 4.2.2. Factor use and structural change The fact that farm profitability did not change much over the period does not exclude that the behav- iour and choices of farmers significantly responded to the change of external conditions (CAP included). In order to more deeply investigate this response is useful to assess whether factor endowment, use and intensities significantly changed within the adopted field of investi- gation. Four fixed (or quasi-fixed) factors are considered: land (UAA); labour (AWU) also including the farm fam- ily labour (FAWU); Machinery (KW); Livestock (LSU) (Sahrbacher et al., 2008). Figure 5 exhibits the evolution of these factors’ endowment over the 2008-2019 period. To facilitate interpretation and comparison, values have been indexed with respect to the initial level (2008=1). For all factors a positive trend can be appreciated whose slope seems to be dependent on the respective degree of fixity. From 2008 to 2019 the average land endowment increased by only 6%, while the growth has been of 10%, 15% and 23% for AWU, LSU and KW, respectively. In fact, live- stock endowment is the only case showing significant 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Coefficient of Variation 3.6 3.5 3.8 3.9 3.4 3.4 3.7 3.4 3.4 3.4 3.3 3.2 Min 0 0 0 0 0 0 0 0 0 0 0 0 1st Quartile 0 0 0 0 0 0 0 0 0 0 0 0 2nd Quartile (Median) 0 0 0 0 0 0 0 0 0 0 0 0 3rd Quartile 0 0 0 0 0 1 0 0 0 0 0 1,401 Max 38,115 51,974 77,603 77,603 77,598 99,500 100,000 73,863 77,341 92,541 92,541 120,010 PILLAR II – OTHERS Mean 1,043 1,010 1,872 1,447 1,408 1,622 1,373 1,690 1,175 1,208 1,980 2,339 Standard deviation 6,853 5,388 8,962 7,567 8,536 8,343 7,605 7,330 5,035 4,720 8,287 7,854 Coefficient of Variation 6.6 5.3 4.8 5.2 6.1 5.1 5.5 4.3 4.3 3.9 4.2 3.4 Min 0 0 0 0 0 0 0 0 0 0 0 0 1st Quartile 0 0 0 0 0 0 0 0 0 0 0 0 2nd Quartile (Median) 0 0 0 0 0 0 0 0 0 0 0 0 3rd Quartile 0 0 0 0 0 0 0 0 0 0 1,047 1,860 Max 184,212 140,000 133,700 149,093 240,000 215,000 240,000 110,000 87,000 83,265 176,513 161,758 % Variation CAP support (2019-2008) Min 1st Quartile 2nd Quartile 3rd Quartile Max -100% -19% +30% +370% - 241The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 oscillations and, more importantly, an apparent trend reversal after 2015. What emerges points to a substantial intensifica- tion in the use of these factors (in fact, the same was observed for the variable inputs). A more detailed analy- sis of the nature of this factors’ intensification is avail- able in the Annex (Table A2). It is worth emphasizing here that, combining the evolution of factors’ use with 0% 5% 10% 15% 20% 25% 0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,000 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 %Environment-using Costs on VC Revenue Variable Costs (VC) Figure 3. Average revenue, variable costs and environment-using costs (fertilizers, pesticides, energy, water) (€) over the 2008-2019 period within the Italian FADN balanced sample. 0% 2% 4% 6% 8% 10% 12% 0 10,000 20,000 30,000 40,000 50,000 60,000 70,000 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 % of farms with negative Net Income - right scale Net Income (nominal terms) - left scale Net Income (real terms, 2010 prices) - left scale Figure 4.Average farm net income (€) over the 2008-2019 period within the Italian FADN balanced sample. 242 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 the profitability dynamics, a decline of factors’ produc- tivity is observed. Figure 6 displays the evolution of the farm net income per unit of labour. Labour productiv- ity (or profitability) declined by 25% from 2008 to 2019, though most of the decline occurs in the very first years of the period. However, if the real term values are con- sidered, the decline is more pronounced (-34%) and occurs quite regularly up to 2014. It is finally interesting to assess whether this evolu- tion in terms of factor endowment, intensities and prof- itability is associated to other structural adjustments concerning farm holders, their turnover and attitudes. Figure 7 reports the presence of female and young (<40 years old) farmers within the sample.19 What emerges is a sharp decline of young holders (from 18% in 2008 to 6% in 2019) and a substantial stability of the presence of female holders (from 15% to 17%). Moreover, there is no 19 It is worth noticing that this sample may significantly underestimate the holders’ turnover. As entry and exit dynamics are excluded by defi- nition within a balanced panel, here only the internal replacements are captured, that is, the possible substitution of the holder within the same farm. Although partial, however, this may still be a reliable representa- tion of the actual structural change occurring within the professional farming sector. Considering agriculture as a whole may misrepresent the presence of female and young farmers as numbers are affected by the presence of very small (non)farms. In the Italian case, in particular, both the presence of female and of elder holders has been always altered by the persistence of these marginal (non)farms (Iacoponi, 2021). Table 2 – Distribution of the farm net income within the Italian 2008-2019 FADN balanced sample (€). 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Mean 57,056 53,945 55,907 55,703 55,380 54,253 51,358 54,100 54,186 55,072 57,512 51,440 Standard deviation 139,843 134,874 135,000 142,913 127,696 119,818 123,255 132,212 133,324 109,159 106,696 106,542 Coefficient of Variation 2.5 2.5 2.4 2.6 2.3 2.2 2.4 2.4 2.5 2.0 1.9 2.1 Min -160,758 -124,741 -143,652 -184,265 -66,484 -40,2051 -18,1687 -165,917 -205,180 -229,603 -121,842 -255,091 1st Quartile 8,969 6,542 9,042 9,147 9,669 9,702 8,423 9,399 9,172 9,573 10,065 8,424 2nd Quartile (Median) 24,802 21,723 25,506 24,741 25,537 25,001 23,146 23,966 24,972 25,785 26,229 23,154 3rd Quartile 58,698 52,296 58,763 57,760 59,681 58,205 53,101 57,595 62,726 6,1431 65,008 58,063 Max 2,429,5722,075,4032,333,8292,228,0931,983,0412,019,8092,100,8503,368,7153,691,6321,815,4411,939,3881,930,918 0.8 0.9 1 1.1 1.2 1.3 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 AWU UAA KW LSU Figure 5. Evolution of main factors’ average endowment (2008=1) over the 2008-2019 period within the Italian FADN balanced sample. 243The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 0 10,000 20,000 30,000 40,000 50,000 60,000 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Net Income/AWU - nominal terms (€) Net Income/FAWU - nominal terms (€) Net Income/AWU - real terms (2010 prices) (€) Net Income/FAWU - real terms (2010 prices) (€) Figure 6. Evolution the farm net income per (F)AWU over the 2008-2019 period within the Italian FADN balanced sample. 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Female - left scale Young - left scale Settlement by succession - right scale Figure 7. Evolution of the presence of the female, young and organic farmers over the 2008-2019 period within the Italian FADN balanced sample. 244 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 evidence of a correspondence between young and female holders as the average age of female and male holders is substantially the same.20 It thus seems difficult to interpret these figures as the progressive emergence of a new generation of farm- ers within the adopted sample. Nonetheless, the share of farmers settled by succession significantly increased from 30% in 2008 to 43% in 2019. This would indicate that 13% of farms experienced a succession during the period of observation. However, this succession is not apparently associated with the takeover of young and female farmers. In addition, most of these successions occurred between 2010 and 2012, thus it may be ques- tioned whether it is real or it is just an artefact due to data collection or some other administrative reason. 4.2.3. Production choices A final aspect of the evolution of farmers’ behav- iour concerns their production choices. The classifica- tion of agricultural holdings by Type of Farming (TF) can be informative in this respect. FADN classifies farms in eight TF categories: five main groups of special- ist agricultural holdings and three mixed groupings.21 Therefore, the first indicator of a production response is expressed by the TF dynamics: a switch from one TF to another evidently expresses the farmer’s decision to change production orientation or specialization. Figure 8 exhibits the evolution of the TF catego- ries over the 2008-2019 period. The most frequent cat- egories are field crops (TF1), permanent crops (TF3) and grazing livestock (TF4). None of the other Types of Farming (TFs) exceeds a 10% share. Overall, shares remain quite constant over time: TF1 remains at 26% even though a slight decline is observed between 2010 and 2016; TF3 remains constant at 30% up to 2014 and then slightly declines to 28%; TF4 starts from 21% and experiences an increase in the first years but then comes back to 22% in 2019. All other TFs show a very limited variation of their share (always lower than 2%). Even the combination of these TFs does not express any significant structural dynamics. For instance, TFs with livestock activities combined (TF4, TF5, TF7 and TF8) show the same share in 2008 and 2019 (31%) with mini- mum changes over the period. 20 See also Giampaolo et al. (2021) and Selmi (2021) for a comparison with analogous evidence on the whole Italian agriculture. 21 The TF of an agricultural holding is determined by the relative impor- tance of each production activity on the total farm SO. The eight groups are defined as follows: TF1 = Field crops; TF2 = Horticulture; TF3 = Permanent crops; TF4 = Grazing livestock; TF5 = Granivores; TF6 = Mixed crops; TF7 = Mixed livestock; TF8 = Mixed crops&livestock. Even though relatively few transitions from one TF to another are observed, it may be interesting to inves- tigate further where these transitions occurs and specu- late on the possible motivations. The Annex (Table A3) provides more details on the observed TF switches. Here, it seems interesting to define the proper dimen- sion of this event. Figure 9 orders the farms per number of TF changes over the 2008-2019 period. For 1079 units (68% of the sample) no change is observed. For other 166 farms (about 10%) only one change is observed. It means that these are genuine switches, namely, in these observations a real change in production orientation has taken place. For all other units, multiple switches are observed during the period. In most cases, they are back-and-forth movements, that is, these farms are momentarily associated to another TF but then go back to the original category. Arguably, this peculiar behav- iour does not express any relevant change in production farmers’ choices. It can be interpreted as physiological oscillations of production activities in borderline farms between two TFs. However, the switch of TF may be a poor indicator of farm production re-orientation. There could be more radical changes in farmer’s output mix that are not cap- tured by the TF classification. It is the case of the acti- vation of unconventional farm activities usually desig- nated as multifunctional diversification: farms combin- ing agricultural production with market or non-market services (multifunctional farms). The FADN dataset provides information about the so-called “Other gainful activities”, also defined as “agriculture-related activities” (“attività connesse”) in Italian regulation.22 Figure 10 displays the evolution of the number of farms with other gainful activities, as well as their inci- dence on the SO both in the whole sample and in these multifunctional farms. For both the number of farms and the incidence on the whole sample, a sharp drop is observed between 2009 and 2010. After that, the trend regularly and consistently reverts to the initial 2008- 2009 variation. It can be argued that this 2009-2010 drop is an artefact due to some changes in data collec- tion as corroborated by the incidence of these activities within these multifunctional farms: it does not show any drop and it increases quite regularly, at least up to 2016. Therefore, if compared to the 2010 level, in 2019 we observe a 3% growth in the number of multifunction- al farms within the sample (from 14% to 17%), a 1.4% growth in the incidence of these activities within the full sample, and a 5% growth in the incidence within mul- 22 They include agritourism and rural tourism, educational farms, active subcontracting, aquaculture, transformation of farm products, produc- tion of renewable energy, environmental services, agro-craft activities. 245The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 tifunctional farms. Therefore, the observed progress of multifunctional activities seems slow overall and it looks like more an increasing specialization of a limited group of farms. Eventually, it appears as a gradual and spon- taneous structural evolution driven more by the market conditions than by some change in the policy support (see below). 5. THE CO-EVOLUTION This section derives from the analysis above some stylised facts about nature and extent of the co-evolution of CAP support and farm behaviour. By co-evolution here we mean that the dynamics of the CAP and the change of farmers’ behaviour concur (so they appear to be correlated) in such a way that it is very difficult, if not unfeasible in practice, to distinguish which is the cause and which is the effect. Therefore, with the term co-evolution we do not want to necessarily mean policy neutrality (or ineffectiveness) in promoting farm practice changes. It may be definitely the case that some agricul- tural practices are triggered by the change in the CAP support. However, empirically assessing this causal link- age, may be very challenging. We want to motivate this conclusion more in detail by separately considering the three abovementioned major policy objectives (income support, production diversification, environmental goods provisions) to which we associate three respective research questions. 5.1. Farm income and CAP support Is there any evidence that CAP payments did really protect the farm’s net income in both level and variabil- 26% 25% 23% 22% 23% 23% 23% 23% 25% 25% 26% 26% 30% 30% 30% 30% 31% 30% 30% 29% 29% 29% 29% 28% 21% 21% 25% 24% 24% 24% 24% 24% 24% 23% 23% 22% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 TF 1 TF 2 TF 3 TF 4 TF 5 TF 6 TF 7 TF 8 Figure 8. Evolution of the Type-of-Farming (TF) categories over the 2008-2019 period within the Italian FADN balanced sample (% is indi- cated only for FT >10%,). Legend: TF1 = Field crops; TF2 = Horticulture; TF3 = Permanent crops; TF4 = Grazing livestock; TF5 = Grani- vores; TF6 = Mixed crops; TF7 = Mixed livestock; TF8 = Mixed crops&livestock. 246 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 10 79 16 6 13 4 73 62 39 16 10 4 1 1 N o change 1 change 2 changes 3 changes 4 changes 5 changes 6 changes 7 changes 8 changes 9 changes 10 changes (6 8% ) 0.0% 0.5% 1.0% 1.5% 2.0% 2.5% 3.0% 3.5% 4.0% 4.5% 0% 5% 10% 15% 20% 25% 30% 35% 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 % of farms with other gainful activities (A) % of other gainful activities on farm SO (only A) % of other gainful activities on farm SO (whole sample) Figure 9. Farms per number of TF changes over the 2008-2019 period within the Italian FADN balanced sample. Figure 10. Evolution of the farms and of the incidence on farm Standard Output of other gainful activities within the Italian FADN bal- anced sample. 247The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 ity? An income-protection effect should imply a nega- tive relationship between the level of CAP support and the farms net income, that is, a larger support for farms showing higher income problems. These problems can be expressed by a negative net income, by a net income that would be negative without the CAP payment (i.e., the ratio between CAP support and net income is >1) or, more generally, by a low labour profitability (i.e., net income per unit of labour). But income problems can be also intended as a large income variability. In order to assess this question, it is worth meas- uring the intensity of support per unit of family labour (FAWU) both to eliminate the size effect and to focus on the actual farmers’ objective variable. Table 3 provides detailed information about the evolution of the CAP support per unit of net income and of AWU and, above all, about its distribution within the sample. Figure 11 displays the CAP support and number of farms with a CAP support larger than the net income (included net income<0). Five major facts are worth noticing. 1. The support per unit of net income significantly oscillates due to the oscillations of the net income itself but, overall, it remains stable over time: 39% in 2008 and 40% in 2019, with a maximum of 66% in 2018 and mini- mum of 24% in 2014.23 2. The support per unit of FAWU increased by 21% in nominal terms (8% in real terms) from 2008 to 2019, but if the comparison is made between 2009 and 2019, the increase falls to 4% in nominal terms and becomes a decline (-7%) in real terms.24 3. The correlation between the CAP support per unit of FAWU and the respective unit net income is signifi- cantly positive25 and it slightly reinforces over time with a maximum of 0.67 in 2018. It indicates that the inci- dence of the CAP support on net income per unit of labour tends to be stronger in farms that need it less as they show an higher labour profitability. 4. The number of farms with a CAP support great- er than net income (negative net income included) is quite stable (around 20%). They receive an almost pro- portional share of support (between 20% and 30%) and the average support to these farms increased by 11% in nominal terms but remained constant in real terms (-0.6%). 23 These figures confirm what emerged in previous studies also for Ital- ian agriculture (European Commission, 2018b). 24 Due the presence of negative values, In computing this indicator, farms with negative net income are attributed the highest incidence observed in the rest of the sample. 25 It is worth reminding that, as detailed in section 4.2.1, the calculation of the net farm income includes the CAP support. Therefore, even when the latter shows a limited incidence on the former on average, a slight positive correlation between the two necessarily occurs. 5. The growth of unit CAP support26 shows a weak but significantly positive correlation with family labour profitability. At the same time, a positive but much stronger correlation is observed between unit support and the variability the family labour profitability. It can be concluded that a quite contradictory evi- dence emerges about the consistency of the CAP as an income protection policy. On the one hand, CAP sup- port may have really supported the farms’ income as its incidence is remarkable. On the other hand, however, support and support growth, though very disperse, go more towards farms that need less, i.e., more profitable farms.27 Therefore, there is no clear indication that this policy is selective in favour of most problematic units but, at the same time, support itself is strongly oriented towards cases showing higher income variability. More than an income support policy, CAP thus seems to behave like an income stabilization policy at whatever income level a farm is. 5.2. Production diversification and CAP support Is there any evidence that the change in CAP pay- ments, either the decoupling of I Pillar payments and the increase of II Pillar payments, induced production diversification? To assess a diversification-inducing effect we need a metric to measure production diversification. Here we firstly follow the analogy with ecological stud- ies where diversity is often measured using the Shannon (or Shannon-Wiener) and the Simpson indexes (Keylock, 2005). These indexes are here adapted to compute the farm-level Diversification Index for any i-th farm at any time t (DIit) (Coderoni, Esposti and Varacca, 2021): (1) Shannon (2) Simpson where c indicates a generic crop/animal species of the set of all observed crops/animal species C. These indexes are separately computed on crops (on the basis of the share on the total farm’s UAA) and on animals (on the basis of the share on the total farm’s LSU), and then averaged weighting by the respective share of crop and livestock products on farm revenue. For both indexes, more diver- sified farms are expected to show an higher DIit and, more 26 For farms with a zero initial CAP support, the attributed growth rate corresponds to observed maximum finite value. 27 A similar evidence for the Italian FADN farms is obtained by Ciliberti et al. (2022). 248 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 importantly, an increased production diversification with- in the sample is expressed by an higher average DIit.28 Figure 12 shows the evolution of the average Shan- non and Simpson diversity indexes within the adopted field of investigation. The two indexes behave similarly though the Shannon index evolves a little more smooth- ly: from 2008 to 2019, the Shannon index increased by 12%, the Simpson index by 10%. As usual, these aver- age values may hide a major heterogeneity within the sample as can be better appreciated by looking at the descriptive statistics reported in Table 4. In both cases, the dispersion (as indicated by the CV) and the asym- metry (as indicated by the median-mean ratio) are lim- 28 The main difference between the two is that the Shannon index rang- es between 0 and lnC/ln2, while Simpson index ranges between 0 and 1. ited compared to most variables investigated above. The growth of the lower quartiles is more intense than the higher ones, thus indicating that not only diversification increased, but also that it distributes more uniformly within the sample. The bottom of Table 4 reports the correlation coef- ficients between these indexes and the CAP support per unit of FAWU. As expected, the two diversity indexes behave very similarly. Therefore, respective results can be commented on together. CAP support by itself shows a little linkage with diversity indexes, at least until 2016 when a positive relationship started to emerge. Appar- ently, this emerging relationship can be attributed to both the II Pillar support and to the I Pillar decoupled support, for which, in fact, the positive linkage emerges from the beginning of the period. Table 3. Evolution of the CAP support per unit of net income and of AWU within the Italian 2008-2019 FADN balanced sample. 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 A) CAP support/Net Income (%) Mean 39% 29% 54% 40% 43% 33% 24% 49% 47% 51% 66% 40% Standard deviation 387% 1,207% 408% 270% 805% 628% 810% 363% 953% 11,234% 732% 266% Coefficient of Variation 9.9 41.6 7.6 6.7 18.9 19.3 34.4 7.3 20.3 220.3 11.1 6.7 Min 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 1st Quartile 0% 0% 3% 4% 5% 4% 4% 3% 2% 3% 4% 5% 2nd Quartile (Median) 17% 21% 24% 24% 24% 22% 25% 25% 22% 22% 24% 25% 3rd Quartile 55% 64% 61% 60% 60% 61% 62% 64% 62% 60% 61% 63% Max 9,868% 16,679% 8,601% 3,553% 21,479% 17,270% 6,455% 9,591% 27,578% 2,388% 23,517% 3,902% B) CAP support/FAWU (€) Mean 13,660 15,954 14,701 14,041 14,789 16,774 16,627 16,239 14,330 15,240 14,561 16,534 Standard deviation 33,102 45,784 40,671 32,212 34,684 53,774 61,150 43,461 36,434 42,160 37,633 39,940 Coefficient of Variation 2.4 2.9 2.8 2.3 2.3 3.2 3.7 2.7 2.5 2.8 2.6 2.4 Min 0 0 0 0 0 0 0 0 0 0 0 0 1st Quartile 556 800 1,096 1,244 1,470 1,330 1,350 1,164 1,043 1,175 1,386 1,754 2nd Quartile (Median) 4,038 4,802 4,519 4,525 5,165 5,369 5,238 5,303 4,414 4,893 4,954 5,592 3rd Quartile 12,498 14,468 13,835 13,489 15,083 14,765 15,201 16,545 14,457 13,841 14,328 16,522 Max 647,037 973,039 781,053 533,976 597,378 1,233,624 1,413,233 1,053,225 883,780 828,248 1,005,035 886,387 Correlation coefficient between B) and net income/ FAWU 0.38* 0.36* 0.60* 0.39* 0.52* 0.64* 0.65* 0.51* 0.50* 0.46* 0.67* 0.54* Correlation coefficient between net income/FAWU and the CAP support 2019-2008 growth rate 0.06* Correlation coefficient between avg. 2019-2008 CAP support/FAWU and standard deviation of net income/FAWU 0.52* a Farms with Net Income<0 are excluded. *Statistically significant at 5% confidence level. 249The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 0% 5% 10% 15% 20% 25% 30% 35% 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 % of farms with CAP>Net Income - right scale % of CAP support to farms with CAP>Net Income on total CAP support - right scale CAP support to farms with CAP>Net Income (000 €) - left scale Figure 11. CAP support and number of farms with CAP support > net income (included net income < 0) within the 2008-2019 Italian FADN balanced sample. 0.35 0.36 0.37 0.38 0.39 0.40 0.41 0.42 1.20 1.25 1.30 1.35 1.40 1.45 1.50 1.55 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Shannon-Wiener - left scale Simpson - right scale Figure 12. Evolution of the average Shannon and Simpson diversity indexes within the Italian 2008-2019 FADN balanced sample. 250 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 A similar analysis can be performed for another set of indicators of production diversification. In this case, it is not an “horizontal” diversification (more crops or livestock activities) but a “vertical” diversification, that is, higher production quality as expressed by process and produc- tion certifications and or by the activation of other gain- ful activities. Table 5 reports the correlation coefficients between CAP support (and its different components) per unit of FAWU and four indicators of this “vertical” diver- sification.29 All these indicators can be expression of a gen- eralized tendency of farmers to look for an improved allo- cation efficiency, i.e., to find the best output mix given the market conditions. In turn, this tendency can be affected by the CAP and its reform in two ways. On the one hand, the progressive decoupling of I Pillar support should ena- 29 Three has to do with certifications: organic farming certification; any kind of environmental certification (organic farming included); any product quality certification but organic certification (for instance, designation of origin). The last indicator is the already discussed multi- functional diversification, that is, the share of other gainful activities on farm’s SO. ble this market reorientation (Esposti, 2017a,b). On the other hand, it can be also the consequence of the II Pillar support itself, as certifications and diversification activities are incentivized by several II Pillar measures. Correlation coefficients reported in Table 5 only weakly support the linkage between the unit CAP sup- port and these diversification indicators. The total CAP support is positively correlated with the organic farm- ing certification (but this linkage is statistically signifi- cant only in the last four years) and negatively correlated with product quality certification. This evidence holds true also for decoupled I Pillar support, while any kind of statistically significant relationship seems to vanish when only coupled I Pillar support is considered. II Pillar unit support shows a very strong positive linkage with organic farming that only slightly weak- ened from 2009 to 2014. A little weaker and more vola- tile, but still positive and mostly statistically significant, is the linkage with all environmental certifications. With only few exceptions concentrated in the initial years of the period, the correlation with II Pillar support statisti- Table 4. Evolution of the Shannon and Simpson diversity indexes within the Italian 2008-2019 FADN balanced sample. 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 A) Shannon diversity index (>1) Mean 1.34 1.35 1.35 1.32 1.33 1.36 1.38 1.42 1.45 1.45 1.47 1.49 Standard deviation 0.94 0.96 0.93 0.94 0.94 0.95 0.97 0.95 0.98 0.98 0.99 1.00 Coefficient of Variation 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 Min 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1st Quartile 0.66 0.66 0.66 0.62 0.64 0.65 0.65 0.72 0.76 0.73 0.74 0.75 2nd Quartile (Median) 1.27 1.31 1.31 1.30 1.31 1.37 1.40 1.44 1.45 1.43 1.44 1.48 3rd Quartile 1.94 1.96 1.96 1.92 1.94 1.99 2.02 2.04 2.09 2.10 2.12 2.14 Max 5.50 5.60 4.85 6.18 5.29 4.97 5.31 5.01 4.80 4.85 5.12 5.50 B) Simpson diversity index (0-1) Mean 0.37 0.38 0.38 0.37 0.38 0.38 0.38 0.40 0.41 0.40 0.40 0.41 Standard deviation 0.26 0.26 0.26 0.26 0.27 0.27 0.27 0.27 0.27 0.27 0.27 0.27 Coefficient of Variation 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.6 Min 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1st Quartile 0.12 0.11 0.11 0.08 0.09 0.09 0.10 0.15 0.16 0.15 0.17 0.17 2nd Quartile (Median) 0.42 0.44 0.44 0.43 0.44 0.44 0.44 0.47 0.47 0.47 0.47 0.47 3rd Quartile 0.60 0.61 0.61 0.60 0.61 0.61 0.61 0.63 0.63 0.63 0.62 0.64 Max 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 Correlation coefficient btw A) and CAP support per FAWU 0.01 -0.02 0.03 0.01 0.00 0.03 0.05 0.04 0.09* 0.05* 0.11* 0.03 Correlation coefficient btw B) and CAP support per FAWU -0.03 -0.03 0.02 -0.03 0.00 0.02 0.03 0.01 0.06* 0.04 0.09* 0.04 Correlation coefficient btw A) and I Pillar decoupled support per FAWU 0.05* 0.03 0.08* 0.07* 0.02 0.03 0.04 0.04 0.08* 0.06* 0.11* 0.04 Correlation coefficient btw B) and I Pillar decoupled support per FAWU 0.01 0.01 0.07* 0.04 0.02 0.01 0.03 0.01 0.07* 0.04 0.10* 0.06* Correlation coefficient btw A) and II Pillar support per FAWU 0.01 -0.02 0.03 0.01 0.00 0.03 0.05 0.04 0.09* 0.05* 0.11* 0.03 Correlation coefficient btw B) and II Pillar support per FAWU -0.01 -0.01 -0.02 0.00 -0.02 0.05* 0.04 0.05* 0.08* 0.02 0.06* 0.02 251The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 cally disappears in the case of product quality certifica- tion and multifunctional diversification. It can be concluded that there is some linkage between the increasing II Pillar support, the progressive decoupling of I Pillar support and production reorienta- tion. However, the empirical evidence is not enough to interpret the observed linkage as an undisputable cause- effect relationship. It can be again interpreted as a co- evolution between market-driven production choices and the path-dependent CAP support.30 30 Its negative linkage with product quality certifications, for instance, can be simply explained by the fact that most of these highly specialised farms were historically recipients of poor support. And of this remains a trace in both decoupled and coupled payments. 5.3. Environmental goods and CAP support Did the change in the CAP support and composi- tion (II Pillar in particular) really induce a greater pro- vision of environmental goods? Also an environmental- good-provision effect of the CAP requires an appropriate metric, i.e., appropriate indicators (Janssen et al., 2010).31 31 This is a challenging task because environmental indicators often require detailed physical information that are hardly available at the farm level and only partially included in the FADN dataset. As part of “the Farm to Fork strategy”, the European Commission has recently announced its intention to convert the FADN into a Farm Sustainabil- ity Data Network (FSDN) to expand the scope of the current FADN network by collecting farm level data also on environmental and social farming practices. Table 5. Correlation coefficients between CAP support per unit of FAWU and different certifications within the Italian 2008-2019 FADN balanced sample. 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Total CAP support/FAWU Organic Farming 0.03 0.01 0.03 0.02 0.00 0.00 0.01 0.02 0.07* 0.09* 0.09* 0.08* Environmental Certification (organic included) 0.01 -0.03 -0.01 0.00 -0.03 0.01 0.00 -0.01 0.02 0.00 0.03 0.04 Product Quality Certification (organic excluded) -0.01 -0.03 -0.06* -0.08* -0.08* -0.07* -0.08* -0.07* -0.07* -0.08* -0.07* -0.06* % of other gainful activities -0.03 -0.01 -0.03 -0.03 0.01 -0.01 -0.03 -0.02 -0.03 -0.02 -0.03 -0.04 Decoupled I Pillar support/FAWU Organic Farming 0.00 0.00 -0.01 -0.01 -0.02 -0.03 -0.02 -0.02 0.05* 0.05* 0.05* 0.02 Environmental Certification (organic included) 0.00 -0.04 -0.03 -0.03 -0.04 -0.02 -0.02 -0.04 -0.03 -0.01 -0.01 0.01 Product Quality Certification (organic excluded) -0.04 -0.04 -0.06* -0.08* -0.08* -0.08* -0.07* -0.08* -0.09* -0.10* -0.08* -0.07* % of other gainful activities -0.04 -0.01 -0.03 -0.03 -0.03 -0.02 -0.03 -0.03 -0.04 -0.04 -0.03 -0.04 Coupled I Pillar support/FAWU Organic Farming -0.02 -0.02 -0.04 -0.04 -0.03 -0.02 -0.01 -0.03 0.00 -0.02 -0.01 -0.03 Environmental Certification (organic included) -0.03 -0.03 -0.04 -0.05* -0.03 -0.01 -0.01 -0.03 0.02 -0.03 -0.03 -0.03 Product Quality Certification (organic excluded) 0.02 -0.03 -0.05* -0.08* -0.03 -0.03 -0.03 0.00 -0.01 -0.03 -0.04 -0.03 % of other gainful activities -0.04 -0.03 -0.03 -0.02 0.01 0.01 -0.01 -0.03 -0.03 -0.02 -0.02 -0.04 II Pillar support/FAWU Organic Farming 0.19* 0.08* 0.14* 0.10* 0.07* 0.10* 0.07* 0.16* 0.20* 0.13* 0.19* 0.18* Environmental Certification (organic included) 0.11* 0.04 0.10* 0.07* 0.03 0.12* 0.05 0.13* 0.17* 0.06* 0.16* 0.12* Product Quality Certification (organic excluded) 0.05* 0.02 -0.01 0.00 0.00 -0.04 -0.03 0.02 0.00 -0.02 0.01 -0.01 % of other gainful activities 0.06* 0.04 0.01 0.00 0.08* 0.01 0.01 0.02 0.01 0.01 0.01 0.02 AEM II Pillar support/FAWU Organic Farming 0.30* 0.17* 0.16* 0.14* 0.15* 0.11* 0.13* 0.19* 0.18* 0.19* 0.20* 0.19* Environmental Certification (organic included) 0.23* 0.08* 0.13* 0.11* 0.08* 0.06* 0.09* 0.14* 0.15* 0.09* 0.14* 0.13* Product Quality Certification (organic excluded) 0.08* 0.02 0.02 0.04 0.02 -0.01 -0.01 0.04 0.02 -0.01 0.02 -0.02 % of other gainful activities 0.03 0.00 -0.02 0.00 0.01 0.00 0.00 -0.01 -0.01 -0.01 0.00 -0.03 Other II Pillar support/FAWU Organic Farming 0.03 -0.01 0.06* 0.04 -0.01 0.03 0.03 0.06* 0.12* 0.04 0.11* 0.10* Environmental Certification (organic included) -0.02 -0.01 0.03 0.02 -0.01 0.12* 0.02 0.06* 0.12* 0.00 0.12* 0.06* Product Quality Certification (organic excluded) 0.00 0.01 -0.02 -0.03 -0.02 -0.05* -0.04 -0.01 -0.02 -0.03 -0.01 0.01 % of other gainful activities 0.06* 0.05* 0.01 0.00 0.10* -0.02 -0.01 0.03 0.01 0.00 -0.01 -0.01 *Statistically significant at 5% confidence level. 252 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 The diversity indexes discussed above may represent proxies of the provision of some environmental services, like the protection of biodiversity within the agro-eco- logical context. But they seem rough indicators of the provision of other environmental goods. At the same time, however, an explicit indication of the achieve- ment of higher environmental standards comes from the abovementioned environmental certifications. Therefore, it is worth investigating further the linkage between these certifications and the CAP support. Figure 13 shows the evolution of the share of farms with organic and environmental certifications. For the sake of comparison, also product quality certifications are reported. All certifications significantly grew over the whole period with +162% for organic farming, +52% for all environmental certifications and +47% for prod- uct quality certifications. In general terms, if we exclude organic farming, environmental certifications seem sub- stantially stagnant compared to product quality certi- fications. Eventually, organic farming has become the prevalent form of environmental certifications over time as it was just 34% on the total in 2008 and reached 58% in 2019. Table 5 presents the correlation coefficients between the two categories of II Pillar CAP support (AEM and other measures) per FAWU and the different certifica- tions. As could be expected, it emerges a strongly posi- tive and significant linkage between AEM payments and environmental certifications, in particular organic farming. On the contrary, there is no evidence of a regu- lar and significant relationship between other II Pillar measures, product quality certifications and multifunc- tional diversification. Even for these measures, the only evidence concerns the linkage with environmental certi- fication, organic farming in particular. It could thus be concluded that a robust relation- ship between the AEM support and organic farming and, more generally, environmental certifications actu- ally emerges. But, again, this does not imply a treat- ment effect as this linkage may be just apparent or, to be more precise, just a tautology. As a matter of fact, certification is not the consequence of a treatment (i.e., a II Pillar measure), but it is the treatment itself: untreated units cannot be certified whereas treated units are automatically certified. Therefore, the TE log- ic might not work properly because the treatment does not leave any behavioural trace, namely, it does not induce any observable behavioural response. In fact, the only behavioural trace is the farmer’s voluntary choice of the treatment itself which inevitably implies certification. 6. CAUSAL INFERENCE, CAP ASSESSMENT AND THE CO-EVOLUTION HYPOTHESIS We can now go back to the original question of the present study, i.e., the actual applicability of the TE logic to CAP assessment. Previous section points to some major features of the co-evolution of CAP support and of farm- ers’ performance. As shown, this co-evolution does not necessarily exclude causation but makes it hardly identi- fiable. In practice, co-evolution is the consequence of the particular forms in which CAP measures are delivered to farmers and these latter progressively take decisions com- bining voluntary participation to these measures with production choices. These forms eventually enter in con- flict with the prerequisites of a TE logic. Without entering into technical details, it must be reminded that almost all CI studies are based on the so-called Potential Outcome (PO) framework (Rubin, 1974; Imbens and Wooldridge, 2009; Imbens and Rubin, 2015). Within this theoretical framework, the empirical identification of the TE depends on the identification of counterfactuals mimicking the outcome variable of a treated unit in the case it was not treated (and the other way round) (Perraillon et al., 2022). But empirical identification and estimation of the TE within this conceptual framework requires an appropriate quasi-experimental design32 and imposes its conditions.33 In particular, six specific sources of conflict between these conditions and the abovementioned forms of co- evolution deserve detailed discussion. Not only they may be all encountered in CAP assessment exercises; more importantly, they may occur simultaneously. Let’s dis- cuss them from the more general (and problematic) to the more technical (and manageable) ones. 6.1. Voluntary and universalistic treatments As discussed at the beginning of this paper, and as shown repeatedly in the empirical analysis, CAP meas- 32 Here we refer to “quasi-experimental design” with the same meaning given by Perraillon et all. (2022) to “research design” on observational units, that is, the overall strategy used to answer a research question with non-experimental data. 33 In particular, three assumptions are critical: the first is the Conditional Independence Assumption (CIA, or Unconfoundedness) that postulates the independence between the potential outcomes and the treatment conditional on a set of pre-treatment (exogenous) variables, or con- founders. The second assumption is the overlap (also known as balance, or positivity, or common support) condition that empirically implies that there must be at least one treated unit and one control unit at each possible value of all confounders. The third condition is the Stable Unit Treatment Value Assumption (SUTVA) that rules out any interference of an individual’s treatment status on another individual’s potential out- come. If these conditions are satisfied, observational data can be regard- ed as generated by a “natural experiment”. 253The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 ures are tendentially universalistic and adoption is most- ly voluntary. All or most farms can apply for these meas- ures and, therefore, the treatment status can not be con- sidered exogenous. This poses fundamental problems in finding suitable counterfactuals as they may not exist at all. Even when non-treated units are present and observ- able, they are so peculiar that can not be confronted with the treated ones: their peculiarity actually is the main reason for their exclusion (either voluntary or not) from the treatment. This makes the application of the TE logic to CAP assessment seriously questionable. Any possible way out of this problem relies more on a prop- er design of the quasi-experimental setting rather than on alternative or adapted TE estimation approaches. In practice, however, available datasets (like the FADN) might make these alternative settings unfeasible. 6.2. Outcome variable The search of an appropriate quasi-experimental set- ting encounters another major issue. It has to do with the ambiguity about the outcome variable to be consid- ered. The empirical analysis here performed clearly illus- trate the point. On the one hand, for many CAP meas- ures a policy target variable is simply neither explicit nor univocal. In such case, the present investigation had to identify, more or less arbitrarily, a suitable metric for the policy assessment. On the other hand, when measures are very clearly targeted (several II Pillar measures, for instance), the outcome variable is clear or univocal but it is just a tautology: the treatment adoption itself implies the outcome variable which automatically takes zero val- ue for the non-treated units. As shown, this is the case, for instance, of certifications’ adoption. An outcome variable may not exist, may be unob- servable, may be multiple or may be tautological. In any case, this poses a fundamental practical challenge for the consistent application of the TE logic to CAP assess- ment. Also in this case, the solution does not necessarily depend on some methodological adaptation or alterna- tive to conventional TE estimation approaches. It rather requires a well suited quasi-experimental design based on a conceptualization of farmers’ behaviour that even- tually leads to the identification of the most appropriate outcome variable to be considered in the analysis.34 34 For a theoretical and empirical investigation on how farmers select the policy and change their behaviour in order to take advantage of it within an utility-maximizing framework, see Esposti (2022). 0% 5% 10% 15% 20% 25% 30% 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Organic Farming Environmental Certification (Organic Farming included) Product Quality Certification (Organic Farming excluded) Figure 13. Evolution of the share of farms with certifications within the Italian 2008-2019 FADN balanced sample. 254 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 6.3. Heterogeneity Coexistence of the voluntaristic and universalistic nature of the CAP aims to cover very diverse farming conditions. As repeatedly emerged in sections 4 and 5, farms under investigation (treated or not) are charac- terized by vast heterogeneity. This has to do with their structural and geographical characteristics, but also with farmer’s personal motivations. While the former features may be observed, the latter remain unobserved and can only be indirectly revealed by the observable farmer’s behaviour (Esposti, 2022). Controlling for this hetero- geneity requires many confounders, thus highly dimen- sional datasets that, in turn, imply remarkable compu- tational complexity (the so-called curse of dimensional- ity). Literature in the field has proposed several solutions (Abadie, 2021) that have also widely adopted in CAP assessment (Esposti, 2017a,b). But farm heterogeneity is challenging also for anoth- er more fundamental, and often disregarded, reason: the TE itself may be strongly heterogeneous. In such case, although the average TE (ATE) is correctly identified and consistently estimated, it simply remains uninforma- tive. Under strong TE heterogeneity, estimating the group or the individual TE is needed for policy assessment and learning (Esposti, 2022). Recently proposed Machine Learning (ML) approaches seems interesting in this respect (Bertoni et al. 2021; Coderoni, Esposti and Varac- ca, 2021; Esposti, 2022). But they are also computationally demanding and complex making their outcome not always transparent and results not fully reliable (Knaus et al., 2021). As a consequence, these approaches also requires a lot of additional validation work (Athey and Imbens, 2017). 6.4. Multivalued treatments Most CI approaches have been designed and applied in a binary treatment context. But, as clearly shown, almost all CAP measures consist in interventions whose intensity varies across discrete or continuous range of possible values (i.e., they are multivalued treatments). A multivalued treatment can be still represented within an augmented PO framework but the empirical implica- tions can be severe. Imbens (2000) and Hirano and Imbens (2004) developed an extension of PO framework to continu- ous multivalued treatments and proposed an estimation approach based on the generalization of the Propensity Score Matching (PSM) of the binary case (Generalised Propensity Score, GPS, estimation) (Esposti, 2017a). However, it provides consistent estimates only whenever the treatment assignment can be considered exogenous once all confounders have been taken into account. The approach proposed by Cerulli (2015) admits this possi- bility of treatment endogeneity and the respective results are consistent even under this circumstance. The application of both approaches, however, may encounter several practical problems for the computa- tional complexity and, above all, for the likely violation of the overlap condition. Alternative non-parametric (or semi-parametric) estimation strategies can be helpful to overcome these issues, but they only apply to discrete (or categorical) multiple treatments (Cattaneo, 2010; Catta- neo et al., 2013; Athey and Imbens, 2017; Esposti, 2017b). Therefore, they may require an arbitrary discretization of continuous treatments. 6.5. Multiple treatments Almost all CI studies concentrate on single treat- ments. As shown, however, the main feature of the CAP and its co-evolution with the farmers’ choices is that it delivers multiple treatments to farms. Identifying and consistently estimating the TE of any single treat- ment with the conventional approaches is possible only under the assumption of treatment independence. But the empirical evidence clarifies that this assumption is quite unrealistic as interdependence is likely to occur both in terms of treatment assignment and in terms of outcome variable. In particular, within the CAP both interdependencies may evidently occur between I and II Pillar measures. In this respect, it could be interesting to assess whether treatments reciprocally interfere by mag- nifying or offsetting the respective TE. At present, how- ever, a viable empirical solution to this issue has not yet emerged (Frolich, 2004; Athey and Imbens, 2017). 6.6. Treatment timing When panel data are available, as in the present case, units can be observed before and after the treat- ment. This allows TE identification and estimation via widely used approaches like the Difference-in-Differ- ences (DID) estimation or the Two-Way Fixed Effects estimation (de Chaisemartin and D’Haultfœuille, 2020). However, though powerful, these approaches still require counterfactuals, with all the abovementioned compli- cations, and imply an additional assumption (the so- called parallel trend assumption) that excludes that time behaves as an additional confounder.35 But what really 35 See Arkhangelsky et al. (2021), Chan and Kwok (2022), Cho et al. (2022), for recent developments in this field. 255The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 makes the timing of the treatment a challenging issue in CAP assessment is that it may differ (in fact, it usu- ally differs) across the treated units. They enter the treat- ment in different moments of time (asynchronous policy adoption). This issue can become even more problematic in the agricultural context as the timing of the farms’ response can be itself heterogenous across units depend- ing on their structural characteristics: even under the same treatment timing, some farms can respond imme- diately others may take some years. Recent generalizations of the DID approach tackle this issue under more than one pre- and post-treatment periods, but still a fixed treatment time (Cerulli, 2019), as well as under many post- and pre-intervention times and with the treatment itself that varies over time (Cerulli and Ventura, 2019; Callaway and Sant’Anna, 2021; Sun and Abraham, 2021). However, as shown in sections 4 and 5, CAP magni- fies these issues and these methodological solutions may be insufficient or unfeasible. Even though, in principle, CAP reforms start at the same time for all farms (at least within a specific EU member state), their actual imple- mentation can differ across space (for instance, regions) and farms may apply in different moments. Moreover, several measures are reiterated across successive CAP programming periods. Consequently, dealing with time- varying treatments is even more challenging because treatment itself may be reiterated on the same units in different periods of time, possibly melded with periods without the treatment. 7. SOME CONCLUDING REMARKS Assessing the farm-level impact of CAP measures and reforms with a TE logic is potentially informative thus highly desirable. Unfortunately, it is also highly challenging. Major theoretical and methodological prob- lems are more often overlooked that explicitly tackled. In this respect, a deeper and more critical discussion within the profession would be desirable. The present paper contributes to this discussion not by proposing an empirical application of methods based on this logic, but presenting an empirical evidence that poses doubts and conditions on their actual applicability. Provided that the target of the policy to be investi- gated is clearly identified (in fact, it is often not clear at all) (Matthews, 2021), empirically assessing whether and to what extent this policy has been successful requires specific pre-conditions. Firstly, we need appropriate datasets. FADN surely is very helpful in this respect, but some of its limitations may reduce the application of these evaluation methodologies. Secondly, and more importantly, we need to investigate the co-evolution of the policy instruments and of the potentially treated units, that is, farmers’ behaviour. Investigating co-evo- lution means finding enough support to the existence of a possible cause-effect relationships and to the feasi- bility of its investigation. In the meaning here given to the term, co-evolution implies that a correlation occurs but this does not necessarily imply causation as it may be the consequence of interdependence between the two processes making an unidirectional cause-effect rela- tionship unidentifiable. On the basis of the empirical investigation here pre- sented and the observed co-evolution, we can conclude that the CAP has really moved in the right direction, that is, consistently with the declared objectives. And the farmers’ changed their behaviour and performance, as well. At the same time, however, this does not mean that the policy induced the expected farmers’ response. Achieving this conclusion within a TE logic requires conditions that are not always compatible with the CAP features. It does not follow that these approaches are and will be always inappropriate in this specific case. It rath- er implies that an acritical adoption of these approach- es may not only lead to wrong policy conclusions but also procrastinates the search for more suited solutions. Moreover, it suggests that any consistent application of these approaches requires more attention on setting up appropriate quasi experimental design with the conse- quent appropriate datasets and theoretical representa- tion of farmers’ choices, and on suitable adaptations and refinements of these approaches. REFERENCES Abadie, A. (2021). Using Synthetic Controls: Feasibil- ity, Data Requirements, and Methodological Aspects. Journal of Economic Literature 59(2): 391–425. Angrist, J.D., Pischke, J-S. (2009). Mostly Harmless Econo- metrics. An Empiricist’s Companion. Princeton Uni- versity Press, Princeton. Anton, J. (2006). Modeling production response to ‘more decoupled’ payments. Journal of Agricultural Interna- tional Trade and Development 2(1): 109–126. ARC2020 (2020). CAP reform post 2020: lost in ambi- tion? Final Report, Agricultural and Rural Conven- tion, Brussels. Arkhangelsky, D., Athey, S., Hirshberg, D.A., Imbens, G.W., Wager, S. (2021). Synthetic Difference-in-Dif- ferences. American Economic Review 111(12): 4088– 4118 256 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 Athey, S., Imbens, G.W. (2017). The State of Applied Econometrics: Causality and Policy Evaluation. Jour- nal of Economic Perspectives 31(2): 3–32. Baldoni, E., Coderoni, S., Esposti, R. (2021). Immigrant workforce and agriculture productivity: Evidence from Italian farm-level data. European Review of Agricultural Economics 48 (4): 805–834. Baldoni, E., Esposti, R. (2021). Agricultural Productiv- ity in Space: an Econometric Assessment Based on Farm-Level Data. American Journal of Agricultural Economics 103(4): 1525-1544. Bertoni, D., Aletti, G., Ferrandi, G., Micheletti, A., Cavic- chioli, D., Pretolani, R. (2018). Farmland use transi- tions after the CAP greening: a preliminary analysis using Markov chains approach. Land Use Policy 79: 789-800. Bertoni, D., Aletti, G., Cavicchioli, D., Micheletti, A., Pre- tolani, R. (2021). Estimating the CAP greening effect by machine learning techniques: A big data ex post analysis. Environmental Science & Policy 119: 44-53. Cagliero, R., Cisilino, F. and Scardera, A. (2010). L’utilizzo Della RICA per la Valutazione di Programmi di Svi- luppo Rurale. Rete Rurale Nazionale, Roma. Callaway, B., Sant’Anna, P.H.C. (2021). Difference-in- Differences with Multiple Time Periods. Journal of Econometrics 225(2): 200-230. Castaño, J., Blanco, M., Martinez, P. (2019). Reviewing Counterfactual Analyses to Assess Impacts of EU Rural Development Programmes: What Lessons Can Be Learned from the 2007–2013 Ex-Post Evaluations? Sustainability 11, 1105: 1-22. Cattaneo, M.D. (2010). Efficient semiparametric estima- tion of multi-valued treatment effects under ignor- ability. Journal of Econometrics 155(2): 138–154. Cattaneo, M.D., Drukker, D.M., Holland, A.D. (2013). Estimation of multivalued treatment effects under conditional independence. The Stata Journal 13(3): 407–450. Cerulli, G. (2015). ctreatreg: command for fitting dose- response models under exogenous and endogenous treatment. The Stata Journal 15(4): 1019–1045. Cerulli, G., Ventura, M. (2019). Estimation of pre- and posttreatment average treatment effects with binary time-varying treatment using Stata. The Stata Journal 19(3): 551-565. Cerulli, G. (2019). TFDIFF: Stata module to compute pre- and post-treatment estimation of the Average Treatment Effect (ATE) with fixed binary treatment. Statistical Software Components, Boston College Department of Economics. Chabé-Ferret, S. and Subervie, J. (2013). How much green for the buck? Estimating additional and wind- fall effects of French agro-environmental schemes by DID-matching. Journal of Environmental Economics and Management 65: 12–27. Chan, M.K., Kwok, S.S. (2022). The PCDID approach: difference-in-differences when trends are potentially unparallel and stochastic. Journal of Business & Eco- nomic Statistics (forthcoming). Cho, R., Desbordes, R., Eberhardt, M. (2022). Too Much Finance... For Whom? The Causal Effects of the Two Faces of Financial Development. CEPR Discussion Paper Series, DP17022, Centre for Economic Policy Research. London. Ciliberti, S., Severini, S., Ranalli, M.G., Biagini, L., Frascarelli, A. (2022). Do direct payments efficiently support incomes of small and large farms? European Review of Agricultural Economics (in press). Coderoni, S., Helming, J., Pérez-Soba, M., Sckokai, P., Varacca, A. (2021). Key policy questions for ex-ate impact assessment of European agricultural and rural policies. Environmental Research Letters 16, 090444. Coderoni, S., Esposti, R., Varacca, A. (2021). How differ- ently do farms respond to agro-environmental poli- cies? A Machine-Learning approach. Department of Agricultural and Food Economics, Università Cat- tolica del Sacro Cuore (Piacenza, Italy), unpublished manuscripts. Collantes, F. (2020). The Political Economy of the Com- mon Agricultural Policy. Coordinated Capitalism or Bureaucratic Monster? Routledge, London. de Chaisemartin, C., D’Haultfœuille, X. (2020). Two- Way Fixed Effects Estimators with Heterogeneous Treatment Effects. American Economic Review 110(9): 2964–2996 Dumangane, M., Freo, M., Granato, S., Lapatinas, A., Maz- zarella, G. (2021). An Evaluation of the CAP impact: a Discrete policy mix analysis. EUR 30880 EN, Publica- tions Office of the European Union, Luxembourg. Ehlers, M-H., Huber, R., Finger, R. (2021). Agricultural policy in the era of digitalization. Food Policy 100, 102019.  Erjavec, E. (2016). Back to the CAP’s future: an inter- est- or evidence-based policy? CAP reform. 1 March 2016. http://capreform.eu/. Accessed on 1st April 2016. Erjavec, K., Erjavec, E. (2015). Greening the CAP’ – Just a fashionable justification? A discourse analysis of the 2014–2020 CAP reform documents. Food Policy 51, 53–62. Esposti, R. (2017a). The empirics of decoupling: Alterna- tive estimation approaches of the farm-level produc- tion response. European Review of Agricultural Eco- nomics 44 (3): 499–537. 257The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 Esposti, R. (2017b). The heterogeneous farm-level impact of the 2005 CAP-first pillar reform: A multivalued treatment effect estimation. Agricultural Economics 48 (3): 373–386. Esposti, R. (2022). Non-Monetary Motivations of Agro- Environmental Policies Adoption. A Causal Forest Approach. Quaderno di Ricerca n. 459, Dipartimento di Scienze Economiche e Sociali, Università Politec- nica delle Marche. Esposti, R., Sotte, F. (2013). Evaluating the Effectiveness of Agricultural and Rural Policies: An Introduction. European Review of Agricultural Economics 40(4): 535-539. European Commission (2018a). Agricultural and farm income. Directorate-Generale for Agriculture and Rural Development, European Commission, Brussels. European Commission (2018b). CAP explained Direct Payments for farmers 2015-2020. Directorate-Gen- erale for Agriculture and Rural Development, Euro- pean Commission, Brussels. European Commission (2019). Analytical factsheet for Italy: Nine objectives for a future Common Agricul- tural Policy. Directorate-Generale for Agriculture and Rural Development, European Commission, Brussels. European Commission (2021). Factsheet – a greener and fairer CAP. Directorate-Generale for Agriculture and Rural Development, European Commission, Brussels. Frascarelli, A. (2020). Direct Payments between Income Support and Public Goods. Italian Review of Agricul- tural Economics 75(3): 25-32. Frascarelli, A. (2021). Direct Payments between Income Support and Public Goods. Italian Review of Agricul- tural Economics 75(3): 25-32. Giampaolo, A., Scardera, A., Carè, V. (2021). Le impren- ditrici agricole. Una fotografia dei dati RICA. Rete Rurale Nazionale, RRN Magazine13: 26-29. Hirano, K., Imbens, G. W. (2004). The propensity score with continuous treatment. In: A. Gelman and X. L. Meng, Applied Bayesian Modeling and Causal Infer- ence from Incomplete-Data Perspectives. West Sussex: Wiley InterScience, 73–84. Iacoponi, V. (2021). L’agricoltura delle donne. Rete Rurale Nazionale, RRN Magazine13: 7-10. Imbens, G.W (2000). The Role of the Propensity Score in Estimating Dose-Response Functions.  Biometrika 87(3): 706–710. Imbens, G.W., Rubin, D.B. (2015). Causal inference in statistics, social, and biomedical sciences. Cambridge University Press. Imbens, G.W., Wooldridge, J.M. (2009). Recent Develop- ments in the Econometrics of Program Evaluation. Journal of Economic Literature 47(1): 5–86. Janssen, S., Louhichi, K., Kanellopoulos, A., Zander, P., Fli- chman, G., Hengsdijk, H., Meuter, E., Andersen, E., Belhouchette, H., Blanco, M. Borkowski, N., Heckelei, T., Hecker, M., Li, H., Oude Lansink, A., Stokstad, G., Thorne, P., van Keulen, H., van Ittersum, M.K., (2010). A Generic Bio-Economic Farm Model for Environ- mental and Economic Assessment of Agricultural Sys- tems. Environmental Management 46: 862–877. Keylock, C.J. (2005), Simpson diversity and the Shannon– Wiener index as special cases of a generalized entro- py. Oikos 109: 203-207. Khagram, S., Thomas, C.W. (2010). Toward a Plati- num Standard for Evidence-Based Assessment. Pub- lic Administration Review. Supplement to Vol. 70: S100-S106. Knaus, M.C., Lechner, M., Strittmatter, A. (2021). Machine learning estimation of heterogeneous causal effects: Empirical Monte Carlo evidence. The Econo- metrics Journal 24 (1): 134–161. Lassance, A. (2020). What Is a Policy and What Is a Government Program? A Simple Question With No Clear Answer, Until Now. Social-Sci- ence Research Network (SSRN), https://ssrn.com/ abstract=3727996  or  http://dx.doi.org/10.2139/ ssrn.3727996. Accessed on 31st December 2021. Latacz-Lohmann, U., Balmann, A., Birner, R., Christen, O., Gauly, M., Grethe, H., Grajewski, R., Martínez, J., Nieberg, H., Pischetsrieder, M., Renner, B., Röder, N., Schmid, J.C., Spiller, A., Taub, F., Voget-Kleschin, L., Weingarten, P. (2019). Designing an effective agri-environment-climate policy as part of the post- 2020 EU Common Agricultural Policy. Berichte über Landwirtschaft, Sonderheft 227. Mack, G., Ferjani, A., Möhring, A., von Ow, A., Mann, S. (2019). How did farmers act? Ex-post validation of linear and positive mathematical programming approaches for farm-level models implemented in an agent-based agricultural sector model. Bio-based and Applied Economics 8(1): 3-19. Mari, F. (2020). The representativeness of the Farm Accounting Data Network (FADN): some sugges- tions for its improvement. Statistical Working Papers, Eurostat, Luxembourg. Matthews, A. (2000). Farm incomes: myths and reality. Cork University Press, Cork, Ireland. Matthews, A. (2021). The contribution of research to agricultural policy in Europe. Bio-based and Applied Economics 10(3): 185-205. OECD (2011). Evaluation of Agricultural Policy Reforms in the European Union. OECD Publishing, Paris. Perraillon, M.C., Lindrooth, R.M., Hedeker, D. (2022). Health Services Research and Program Evaluation: 258 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 Causal Inference and Estimation. Cambridge Univer- sity Press (forthcoming). Pupo D’Andrea, M.R. (2021). Le novità della PAC 2023- 2027. Agriregionieuropa, Numero Speciale - Agri- calabriaeuropa n. 1, 2-6. Rocchi, B., Stefani, G., Romano, D. (2012). Differenze di reddito tra famiglie agricole e non agricole in Italia: una verifica empirica. Agriregionieruopa 31: 73-76. Rubin, D. B. (1974). Estimating Causal Effects of Treat- ments in Randomized and Nonrandomized Studies. Journal of Educational Psychology 66(5): 688–701. Sahrbacher, C., Kellermann, K., Balmann, A. (2008). Winners and Losers of Policy Changes – What Is the Role of Structural Change? Paper presented at the 107th EAAE Seminar, European Association of Agri- cultural Economists, January 30-February 1, Sevilla, Spain. Selmi, U. (2021). Parità di genere e agricoltura, il modello è la multifunzionalità. Rete Rurale Nazionale, RRN Magazine 13: 17-20. Sotte, F. (2006). Imprese e non-imprese nell’agricoltura Italiana. Politica Agricola Internazionale 1/2006: 13–30. Sotte, F. (2014). La geografia della nuova PAC in Italia. Agriregionieuropa 38: 11-15. Sotte, F. (2021a). Riflessioni sulla futura politica agricola europea. Agriregionieuropa, Numero Speciale - Agri- calabriaeuropa n. 1, 7-11. Sotte, F. (2021b). La politica agricola europea. Storia e analisi.  Collana Economia Applicata, Agriregion- ieuropa - Associazione Alessandro Bartola, Ancona. Sotte, F. and Arzeni, A. (2013). Imprese e non-imprese nell’agricoltura italiana. AgriRegioniEuropa 32: 65–70. Sun, L., Abraham, S. (2021). Estimating dynamic treat- ment effects in event studies with heterogeneous treatment effects. Journal of Econometrics 225: 175– 199. Swinnen, J. (ed.) (2015). The Political Economy of the 2014-2020 Common Agricultural Policy. An Imperfect Storm. Rowman & Littlefield Publishers/Center for European Policy Studies, London. Terluin, I., Verhoog, D. (2018). Distribution of CAP pillar 1 payments to farmers in the EU. Wageningen Uni- versity & Research, Wageningen Economic Research report 2018-039b. Vrolijk, H., Poppe, K. (2021). Cost of Extending the Farm Accountancy Data Network to the Farm Sustainabil- ity Data Network: Empirical Evidence.  Sustainabil- ity 13, 8181. ANNEX A1. Evolution of the AEM support Figure A1 shows how AEM payments evolved in terms of number of beneficiaries and of average sup- port per beneficiary. The growth of AEM support comes from the combination of two facts. On the one hand, the number of beneficiaries increased by 75% passing from 245 farms (15% of the whole balanced sample in 2008) to 428 units (27% in 2019). On the other hand, the aver- age payment per farm increased almost with the same intensity (+70%) passing from about 5.3 thousand € in 2008 to 9.1 thousand € in 2019. In fact, the growth of the number of beneficiaries is not regular as it shows a fall from 2012 to 2015 and, then, a jump as a consequence of the transition from one regime to another. This sort of bureaucratic cycle is somehow compensated by the countermovement of the average payment per farm that reaches its peak exactly in 2015. A2. The Lorentz curve of the farms’ CAP support and income To better illustrate the distributional characteristics of CAP support, and its evolution over time, within the sample, the Lorentz curves of the Pillar I and Pillar II support, respectively, are reported in Figure A2 for years 2008, 2015, 2019. The sharp concentration of the support on a very limited number of farms clearly emerges. As expected, it is higher in the case of II Pillar where 5% and 3% of farms (i.e., 79 and 48 farms) concentrate 50% of the support in 2019 and in 2008, respectively. But this over concentration is only a little lower for I Pillar with 8% and 6% (127 and 95 farms), respectively. Within the adopted field of investigation, the sequence of CAP reforms has slightly changed the distribution of the CAP support by making it a little bit more homogenous. But this change remains almost negligible. Figure A3 presents the analogous Lorentz curves of the farm net income for selected years 2008, 2015 and 2019.36 Two aspects are worth noticing. First, as expect- ed, the distribution of net income within the sample is highly asymmetric with very few units concentrat- ing most of the total (positive) net income. Second, no significant change in this distribution can be appreci- ated moving from 2008 to 2019. Eventually, in 2008 9% of farms concentrated 50% of the total (positive) net income; in 2019, this share has slightly increased to 11%. 36 These curves are obtained considering only farms with a positive net income in the respective year. 259The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 A3. Factors’ intensification To better investigate the nature of factors’ inten- sification, Table A2 reports the distributional charac- teristics of the factor intensities per labour unit (AWU) together with labour profitability. It firstly emerges that these structural characteristics remain quite stable over time as could be expected considering that adjustments in (quasi)fixed factors’ endowment take time and may have a cost (Esposti, 2017a). It emerges a small reduc- tion in the incidence of family labour on the total farm’s labour use (-3.2%). Also the land endowment per unit of labour slightly declines (-4.8%). But for the other pro- duction factors, it emerges a gradual intensification with a 11% increase of machinery endowment, a 8% increase of the livestock endowment and, above all, a 18% increase of environment-using costs per unit of labour. Although these ratios should get rid of the size effect, with the only exception of the FAWU/AWU ratio, they show a remarkable heterogeneity. Also for these struc- tural characteristics and their evolution, a major disper- sion (as expressed by CV) and asymmetry (as expressed by the median/mean ratio) emerges within the field of investigation. For instance, in the case of land endow- ment, we range from no-land farms to observations with hundreds of hectares per unit of labour. The bottom line of this large heterogeneity is expressed by the net income per unit of labour reported in the final rows of Table A2. Here we also find negative values and this makes the dispersion even more evident. Values range from a minimum of -345 thousand € per unit of labour in 2008 to a maximum 2372 thousand € per unit of labour in 2009. Only a little decline of dispersion of asymmetry is observed in the post 2015 period. More importantly, the mean value significantly declines over the 2008-2019 period (-13% in nominal terms; -22% in real terms) and this reveals a significant redistribution in favour of the more profitable farms: while 1st and 2nd quartiles decline by 15% and 20% respectively, the 3rd quartile declines by only 6% and the maximum value increases by 8%. A4. TF switches In order to only focus on real changes in production orientation, we limit our attention to those switches that make the initial TF of farm differ from the final one. These switches concern 187 farms (12% of the sample). These movements are positioned in a Source-Destina- tion matrix by TF category (Table A3).37 As could be 37 Therefore, the diagonal elements indicate the non-switching units. 0 50 100 150 200 250 300 350 400 450 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000 10,000 11,000 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Number of farms - right scale Total support (000 €) - left scale Avg. per farm (€) - left scale Figure A1. Evolution of the Agro-Environmental Measures (AEM) support within the Italian 2008-2019 FADN balanced sample: number of beneficiaries, total support and average support per beneficiary. 260 Roberto Esposti Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2 27 52 77 10 2 12 7 15 2 17 7 20 2 22 7 25 2 27 7 30 2 32 7 35 2 37 7 40 2 42 7 45 2 47 7 50 2 52 7 55 2 57 7 60 2 62 7 65 2 67 7 70 2 72 7 75 2 77 7 80 2 82 7 85 2 87 7 90 2 92 7 95 2 97 7 10 02 10 27 10 52 10 77 11 02 11 27 11 52 11 77 12 02 12 27 12 52 12 77 13 02 13 27 13 52 13 77 14 02 14 27 14 52 14 77 15 02 15 27 15 52 15 77 2008 2015 2019 a) 92% 94% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1 26 51 76 10 1 12 6 15 1 17 6 20 1 22 6 25 1 27 6 30 1 32 6 35 1 37 6 40 1 42 6 45 1 47 6 50 1 52 6 55 1 57 6 60 1 62 6 65 1 67 6 70 1 72 6 75 1 77 6 80 1 82 6 85 1 87 6 90 1 92 6 95 1 97 6 10 01 10 26 10 51 10 76 11 01 11 26 11 51 11 76 12 01 12 26 12 51 12 76 13 01 13 26 13 51 13 76 14 01 14 26 14 51 14 76 15 01 15 26 15 51 15 76 2008 2015 2019 b) 95% 97% Figure A2 – Lorentz curves of the Pillar I (a) and Pillar II (b) support within the Italian 2008-2019 FADN balanced sample: years 2008, 2015, 2019. 261The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 expected, flows mostly concern two kind of movements: one occur across the main TFs, (TF1, TF3 and TF4); the other concerns movements from more specialized TFs to the mixed ones (TF6, TF7 and TF8). Nonetheless, no prevalent migration emerges and this confirms that, over the period of observation, there is no prevalent evolu- tionary dynamic expressing a generalised reorientation of the farmers’ production choices. 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1 26 51 76 10 1 12 6 15 1 17 6 20 1 22 6 25 1 27 6 30 1 32 6 35 1 37 6 40 1 42 6 45 1 47 6 50 1 52 6 55 1 57 6 60 1 62 6 65 1 67 6 70 1 72 6 75 1 77 6 80 1 82 6 85 1 87 6 90 1 92 6 95 1 97 6 10 01 10 26 10 51 10 76 11 01 11 26 11 51 11 76 12 01 12 26 12 51 12 76 13 01 13 26 13 51 13 76 14 01 14 26 14 51 2008 2015 2019 89%91% Figure A3. Lorentz curve of the (positive) farm net income within the Italian 2008-2019 FADN balanced sample: years 2008, 2015, 2019. 262 Antonio Lopolito1,*, Angela Barbuto2, Fabio Gaetano Santeramo2 Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 Table A1. Representativeness of the balanced FADN sample. Comparison of the Italian 2008-2019 FADN balanced panel (year 2010) with the Italian 2010 agricultural Census: distribution by Types of Farming (TF) and Economic Size (ES) classes (SO=Standard Output). FADN balanced sample 2010 Census (Total) 2010 Census (SO>8000 €) TF classes: TF 1 23% 24% 23% TF 2 8% 2% 6% TF 3 30% 55% 43% TF 4 25% 8% 16% TF 5 3% 1% 1% TF 6 6% 7% 7% TF 7 1% 0% 1% TF 8 4% 2% 4% Not Classified 0% 1% 0% Total 100% 100% 100% ES classes: Small (SO <25,000 €) 30% 18% 49% Medium-Small (SO=25,000-50,000 €) 19% 8% 21% Medium (SO=50,000-100,000 €) 22% 5% 15% Medium-Large (SO=100,000-250,000 €) 20% 4% 10% Large (SO>250,000 €) 9% 2% 5% Total 100% 37% 100% Legend: TF1 = Field crops; TF2 = Horticulture; TF3 = Permanent crops; TF4 = Grazing livestock; TF5 = Granivores; TF6 = Mixed crops; TF7 = Mixed livestock; TF8 = Mixed crops&livestock. Source: FADN and ISTAT. Table A2.Factor use and profitability per labour unit within the Italian 2008-2019 FADN balanced sample. 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Family AWU/AWU Mean 0.71 0.68 0.73 0.71 0.70 0.70 0.70 0.69 0.73 0.71 0.72 0.69 Standard deviation 0.31 0.31 0.29 0.30 0.30 0.29 0.29 0.30 0.30 0.30 0.29 0.31 Coefficient of Variation 0.43 0.46 0.39 0.42 0.43 0.42 0.42 0.43 0.41 0.43 0.40 0.45 Min 0.00 0.00 0.00 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.02 0.00 1st Quartile 0.43 0.40 0.48 0.46 0.44 0.46 0.45 0.43 0.49 0.44 0.48 0.41 2nd Quartile (Median) 0.80 0.72 0.79 0.80 0.76 0.73 0.74 0.74 0.85 0.79 0.81 0.73 3rd Quartile 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 Max 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 UAA/AWU (ha) Mean 18.5 18.5 17.8 17.4 17.4 17.3 17.5 18.3 17.7 17.8 18.1 17.6 Standard deviation 26.6 25.6 24.3 23.7 24.4 23.9 24.4 31.5 22.9 22.9 23.2 21.9 Coefficient of Variation 1.44 1.39 1.36 1.36 1.40 1.38 1.39 1.72 1.29 1.28 1.28 1.24 Min 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.0 0.1 0.1 0.1 1st Quartile 3.8 3.9 3.9 4.0 3.9 3.9 3.9 4.0 4.0 4.1 4.0 4.0 2nd Quartile (Median) 9.5 9.6 9.6 9.4 9.4 9.2 9.2 9.6 9.7 9.3 9.6 9.7 3rd Quartile 23.0 22.9 22.3 22.4 21.8 21.7 21.6 21.6 22.3 22.2 22.1 21.9 Max 486.4 387.3 387.3 421.8 421.8 421.8 387.3 803.6 274.5 200.5 192.4 179.1 263The role of network characteristics of the innovation spreaders in agriculture Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 KW/AWU (hp) Mean 113.1 112.4 112.2 112.7 114.0 114.0 116.4 124.0 120.3 123.0 123.7 125.7 Standard deviation 117.1 104.8 110.3 100.7 98.2 98.1 100.3 241.8 109.4 115.8 117.3 121.4 Coefficient of Variation 1.03 0.93 0.98 0.89 0.86 0.86 0.86 1.95 0.91 0.94 0.95 0.97 Min 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1st Quartile 45.3 47.7 50.0 50.2 51.8 50.9 51.5 51.8 52.2 51.2 53.0 53.6 2nd Quartile (Median) 80.8 82.9 82.1 82.8 86.1 87.3 88.0 90.0 92.4 90.3 91.2 90.6 3rd Quartile 137.8 143.6 143.5 145.9 143.7 144.3 148.1 148.2 151.9 155.6 155.4 155.5 Max 1,341 1,010 2,010 812 798 926 846 8,560 1,488 1,123 1,488 1,488 LSU/AWU Mean 12.3 12.4 14.2 14.3 13.7 15.1 14.4 14.5 15.8 14.8 13.6 13.2 Standard deviation 36.4 41.3 44.5 57.8 40.7 55.1 46.8 56.3 62.4 56.2 43.6 46.9 Coefficient of Variation 2.97 3.32 3.13 4.03 2.98 3.65 3.25 3.87 3.95 3.81 3.21 3.54 Min 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1st Quartile 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 2nd Quartile (Median) 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 3rd Quartile 11.1 11.9 13.0 12.7 12.5 12.3 13.0 12.0 11.5 10.6 10.1 9.4 Max 528 1,032 992 1,782 580 881 860 1,042 1,125 1,291 604 723 Environment-using Costs/AWU (€) Mean 5,498 5,432 5,501 5,920 6,197 6,033 6,223 6,908 6,419 6,602 6,398 6,488 Standard deviation 8,373 7,237 8,036 7,993 8,199 7,813 8,101 19,089 9,226 11,042 9,776 9,830 Coefficient of Variation 1.52 1.33 1.46 1.35 1.32 1.30 1.30 2.76 1.44 1.67 1.53 1.52 Min 0 0 0 0 0 0 0 0 0 0 0 0 1st Quartile 1,452 1,444 1,543 1,737 1,856 1,827 1,938 1,873 1,669 1,688 1,670 1,705 2nd Quartile (Median) 3,068 3,184 3,199 3,484 3,598 3,634 3,743 3,678 3,602 3,608 3,691 3,764 3rd Quartile 5,940 6,072 5,957 6,428 6,838 6,635 6,914 6,960 7,002 6,927 7,031 7,018 Max 102,031 64,425 84,848 75,441 92,735 73,174 82,336 671,360 119,471 189,599 154,697 132,348 Net Income/Family AWU (€) Mean 44,928 50,549 43,592 45,238 45,209 45,561 43,007 45,313 43,174 45,413 45,861 45,113 Standard deviation 103,713 141,432 107,183 124,906 102,187 115,314 126,977 104,157 107,065 95,329 99,178 97,087 Coefficient of Variation 2.31 2.80 2.46 2.76 2.26 2.53 2.95 2.30 2.48 2.10 2.16 2.15 Min -456,321 -166,321 -82,087 -80,300 -64,492 -181,104 -182,261 -162,664 -170,206 -229,603 -69,855 -208,142 1st Quartile 5,919 4,881 6,080 6,016 7,051 6,647 5,871 7,063 6,116 7,134 6,955 5,817 2nd Quartile (Median) 18,756 16,773 18,025 17,781 19,567 18,784 16,593 19,057 17,168 18,894 19,262 17,367 3rd Quartile 45,051 45,133 43,189 43,864 45,927 46,124 43,316 46,335 46,287 49,011 48,762 48,544 Max 1,454,834 3,459,005 1,944,858 2,197,699 1,246,851 2,693,079 3,166,903 2,041,645 1,986,362 1,609,615 2,085,363 1,821,656 Net Income/AWU (€) Mean 33,991 34,658 33,194 33,845 31,891 31,971 29,003 29,232 31,445 30,749 32,729 29,628 Standard deviation 78,467 96,969 81,619 93,450 72,083 80,916 85,630 67,193 77,978 64,547 70,778 63,762 Coefficient of Variation 2.31 2.80 2.46 2.76 2.26 2.53 2.95 2.30 2.48 2.10 2.16 2.15 Min -345,243 -114,033 -62,508 -60,077 -45,493 -127,082 -122,912 -104,936 -123,965 -155,465 -49,852 -136,698 1st Quartile 4,479 3,346 4,630 4,501 4,974 4,664 3,959 4,556 4,455 4,831 4,963 3,820 2nd Quartile (Median) 14,190 11,500 13,726 13,303 13,803 13,181 11,190 12,294 12,503 12,793 13,746 11,406 3rd Quartile 34,085 30,944 32,888 32,817 32,397 32,365 29,211 29,891 33,712 33,185 34,799 31,881 Max 1,100,695 2,371,566 1,480,981 1,644,227 879,534 1,889,753 2,135,670 1,317,088 1,446,709 1,089,877 1,488,212 1,196,382 264 Antonio Lopolito1,*, Angela Barbuto2, Fabio Gaetano Santeramo2 Bio-based and Applied Economics 11(3): 231-264, 2022 | e-ISSN 2280-6172 | DOI: 10.36253/bae-12912 Table A3. Source-Destination matrix for TF category within the Italian 2008-2019 FADN balanced sample (in grey >10 elements). Destination Source TF 1 TF 2 TF 3 TF 4 TF 5 TF 6 TF 7 TF 8 Total TF 1 294 10 20 35 6 57 3 36 460 TF 2 10 96 22 0 0 14 0 0 142 TF 3 20 10 375 4 2 47 2 12 471 TF 4 33 0 5 272 2 6 17 46 382 TF 5 5 0 3 2 28 2 3 5 48 TF 6 55 6 58 5 3 9 0 5 141 TF 7 2 0 2 13 4 0 0 0 22 TF 8 35 0 15 36 7 6 2 5 105 Total 455 122 501 366 52 141 26 110 1772 Legend: TF1 = Field crops; TF2 = Horticulture; TF3 = Permanent crops; TF4 = Grazing livestock; TF5 = Granivores; TF6 = Mixed crops; TF7 = Mixed livestock; TF8 = Mixed crops&livestock. Volume 11, Issue 3 - 2022 Firenze University Press Bio-based Business Models: specific and general learnings from recent good practice cases in different business sectors Nora Hatvani1,*, Martien J.A. van den Oever2, Kornel Mateffy1, Akos Koos1 Food loss and waste accounting: the case of the Philippine food supply chain Anieluz Pastolero*, Maria Sassi The role of network characteristics of the innovation spreaders in agriculture Antonio Lopolito1,*, Angela Barbuto2, Fabio Gaetano Santeramo2 The co-evolution of policy support and farmers behaviour. An investigation on Italian agriculture over the 2008-2019 period Roberto Esposti Price dependence of biofuels and agricultural products on selected examples Wioleta Sobczak*, Jarosław Gołębiewski